A method of igniting a fuel-air mixture with frequency-stabilized energy

Abstract

This invention discloses a frequency- and energy-stable ignition method. Step S1: Collect historical data on environmental parameters, feedback signals from the inverter and discharge circuits, energy storage parameters, discharge excitation signals, and the output voltage of the secondary boost circuit; and label the historical data as training samples. Step S2: Construct a parameter prediction model based on a neural network to predict the feedback signals from the output inverter and discharge circuits, as well as the discharge excitation signal; train the parameter prediction model using the training samples. Step S3: Use the trained parameter prediction model to predict the control parameters of the inverter control circuit and the discharge control circuit; based on the prediction results, construct the excitation function using a reinforcement learning algorithm, and iteratively optimize to obtain the optimal control parameters. This invention optimizes the control parameters of the inverter control circuit and the discharge control circuit through a deep learning parameter prediction model, achieving optimal adjustment of the ignition circuit's frequency and energy, and possesses good practicality.

Description

Technical Field

[0001] This invention belongs to the technical field of ignition devices for aero-engines, and specifically relates to an ignition method with stable frequency and energy. Background Technology

[0002] Current common gas engine ignition methods, limited by mechanical oscillator inverters and gas discharge tubes, are significantly affected by external environmental factors, resulting in large variations in spark frequency and energy storage. This cannot guarantee the success rate of starting and ignition within the specified task profile, thus delaying or affecting the completion of the gas engine's tasks. The stable energy and frequency ignition method in this invention controls the spark frequency through an inverter control circuit and the energy storage through a discharge control circuit. This eliminates the influence of changes in input power and environmental conditions on frequency and energy, which not only improves the lifespan, reliability, voltage fluctuation resistance, and temperature resistance of the ignition system, but also reduces the design complexity of the engine combustion chamber and starting system, improving the reliability and stability of engine starting. Summary of the Invention

[0003] The purpose of this invention is to provide an ignition method with stable frequency and energy. By optimizing the control parameters of the inverter control circuit and the discharge control circuit through a parameter prediction model of deep learning, the frequency and energy of the ignition circuit can be optimally adjusted, which has good practicality.

[0004] This invention is mainly achieved through the following technical solutions:

[0005] A frequency- and energy-stable ignition method employs an ignition system comprising, in sequence, an inverter control circuit, an inverter circuit, a rectifier energy storage circuit, a discharge circuit, a secondary boost circuit, and an ignition nozzle. A discharge control circuit is provided between the rectifier energy storage circuit and the discharge circuit. The inverter control circuit receives feedback signals from the inverter circuit and the discharge circuit and controls the inverter frequency of the inverter circuit. The method includes the following steps:

[0006] Step S1: Collect historical data on environmental parameters, feedback signals from the inverter circuit and discharge circuit, energy storage parameters, discharge excitation signal, and output voltage of the secondary boost circuit; and mark the historical data as training samples.

[0007] Step S2: Construct a parameter prediction model based on a neural network to predict the feedback signals of the output inverter circuit and discharge circuit, as well as the discharge excitation signal; train the parameter prediction model using training samples.

[0008] Step S3: Use the trained parameter prediction model to predict the control parameters of the inverter control circuit and the discharge control circuit. Based on the prediction results, use a reinforcement learning algorithm to construct the excitation function and iteratively optimize to obtain the optimal control parameters.

[0009] To better realize the present invention, the parameter prediction model further includes a frequency feature network, a voltage feature network, an environmental feature network, and a feature fusion layer in parallel, wherein the feature fusion layer is connected to the frequency feature network, the voltage feature network, and the environmental feature network respectively; the frequency feature network, the voltage feature network, and the environmental feature network are used to extract feedback signals, discharge excitation signals, and environmental parameters respectively.

[0010] To better realize the present invention, the parameter prediction model further includes an attention layer, and the front end of the feature fusion layer is provided with an attention layer, which is connected to the frequency feature network, the voltage feature network and the environmental feature network respectively.

[0011] To better realize the present invention, step S1 further includes preprocessing the data to fill in missing values, delete outliers and duplicate values.

[0012] To better implement the present invention, in step S1, the data is first completed and then normalized.

[0013] To better realize the present invention, the environmental parameters further include temperature and humidity.

[0014] The beneficial effects of this invention are as follows:

[0015] This invention optimizes the control parameters of the inverter control circuit and the discharge control circuit using a parameter prediction model based on deep learning. It considers the combined effects of environmental parameters, feedback signals from the inverter and discharge circuits, energy storage parameters, and discharge excitation signals on the overall circuit control parameters. The excitation function of the reinforcement learning algorithm is used to select the optimal combination of control parameters, thereby achieving optimal adjustment of the frequency and energy of the ignition circuit, effectively ensuring ignition quality, and demonstrating good practicality. Attached Figure Description

[0016] Figure 1 This is a block diagram of the ignition circuit principle of the present invention.

[0017] Among them: 1-inverter circuit, 2-rectifier energy storage circuit, 3-discharge circuit, 4-secondary boost circuit, 5-inverter control circuit, 6-discharge control circuit, 7-ignition nozzle. Detailed Implementation

[0018] Example 1:

[0019] A frequency- and energy-stable ignition method employs an ignition system comprising, in sequence, an inverter control circuit 5, an inverter circuit 1, a rectifier energy storage circuit 2, a discharge circuit 3, a secondary boost circuit 4, and an ignition nozzle 7. A discharge control circuit 6 is provided between the rectifier energy storage circuit 2 and the discharge circuit 3. The inverter control circuit 5 receives feedback signals from the inverter circuit 1 and the discharge circuit 3 and controls the inverter frequency of the inverter circuit 1. The method includes the following steps:

[0020] Step S1: Collect historical data of environmental parameters, feedback signals of inverter circuit 1 and discharge circuit 3, energy storage parameters, discharge excitation signal and output voltage of secondary boost circuit 4; and mark the historical data as training samples;

[0021] Step S2: Construct a parameter prediction model based on a neural network to predict the feedback signals of the output inverter circuit 1 and the discharge circuit 3, as well as the discharge excitation signal; train the parameter prediction model using training samples.

[0022] Step S3: Use the trained parameter prediction model to predict the control parameters of the inverter control circuit 5 and the discharge control circuit 6. Based on the prediction results, use a reinforcement learning algorithm to construct the excitation function and iteratively optimize to obtain the optimal control parameters.

[0023] Preferably, the parameter prediction model includes a frequency feature network, a voltage feature network, an environmental feature network, an attention layer, and a feature fusion layer in parallel, wherein the attention layer is connected to the frequency feature network, the voltage feature network, and the environmental feature network, respectively; the frequency feature network, the voltage feature network, and the environmental feature network are used to extract feedback signals, discharge excitation signals, and environmental parameters, respectively.

[0024] Preferably, step S1 further includes preprocessing the data, first filling in missing values, deleting outliers and duplicate values, and then performing normalization.

[0025] The parameter prediction model integrates information extracted from different influencing factors through parallel frequency feature networks, voltage feature networks, and environmental feature networks into a stable multi-factor representation. An attention mechanism is applied to the fusion operation of each network. This attention mechanism typically refers to the weighted sum of a set of scalar weight vectors dynamically generated by a set of "attention" models at each time step. The multiple output heads of this attention mechanism can dynamically generate the weights used in the summation, thus preserving additional weight information during the final concatenation.

[0026] Deep reinforcement learning can directly select actions based on raw input data, making it an artificial intelligence algorithm that more closely resembles human thinking. Deep learning approximates functions by learning deep, non-linear network structures and the essential characteristics of the dataset. In the process of interacting with the environment, the agent uses reinforcement learning to generate optimal behavioral strategies through continuous trial and error and maximizing cumulative rewards.

[0027] Preferably, such as Figure 1 As shown, inverter circuit 1 inverts (10~30)V low-voltage DC power into high-voltage pulse power, which is then fed into rectifier energy storage circuit 2 to store electrical energy in energy storage capacitor. Discharge control circuit 6 generates discharge excitation signal to discharge circuit 3 based on the energy storage status of rectifier energy storage circuit 2, controlling the working state of discharge circuit 3. Discharge circuit 3 then transmits the energy stored in energy storage capacitor to secondary boost circuit 4. Secondary boost circuit 4 raises the output voltage to the voltage required for normal operation of ignition nozzle 7, turns on ignition nozzle 7, and forms an electric spark at its ignition end.

[0028] During operation, the inverter control circuit 5 receives feedback signals from the inverter circuit 1 and the discharge circuit 3, and inputs the inverter control signal to the inverter circuit 1 to control its inverter frequency. The MOSFETs contained in the inverter circuit 1 serve as inverter switching devices. The inverter control circuit 5 uses a PWM chip to stabilize the spark frequency and energy of the ignition device by adjusting the oscillation frequency, setting the inverter current, and comparing the discharge feedback voltage.

[0029] This invention optimizes the control parameters of the inverter control circuit 5 and the discharge control circuit 6 using a parameter prediction model based on deep learning. It considers the combined effects of environmental parameters, feedback signals from the inverter circuit 1 and the discharge circuit 3, energy storage parameters, and discharge excitation signals on the overall circuit control parameters. The excitation function of the reinforcement learning algorithm is used to select the optimal combination of control parameters, thereby achieving optimal adjustment of the frequency and energy of the ignition circuit, effectively ensuring ignition quality, and demonstrating good practicality.

[0030] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A frequency-stable and energy-stable ignition method, which is carried out by using an ignition system, the ignition system comprising, in sequence, an inverter control circuit (5), an inverter circuit (1), a rectification energy storage circuit (2), a discharge circuit (3), a secondary voltage-boosting circuit (4) and an ignition electrode (7), and a discharge control circuit (6) is arranged between the rectification energy storage circuit (2) and the discharge circuit (3), the inverter control circuit (5) is used for receiving feedback signals of the inverter circuit (1) and the discharge circuit (3) and controlling the inverter frequency of the inverter circuit (1); characterized in that, Includes the following steps: Step S1: Collect historical data of environmental parameters, feedback signals of inverter circuit (1) and discharge circuit (3), energy storage parameters, discharge excitation signal and output voltage of secondary boost circuit (4); and mark the historical data as training samples; Step S2: Construct a parameter prediction model based on a neural network to predict the feedback signals of the output inverter circuit (1) and the discharge circuit (3) as well as the discharge excitation signal; train the parameter prediction model using training samples; Step S3: Use the trained parameter prediction model to predict the control parameters of the inverter control circuit (5) and the discharge control circuit (6). Based on the prediction results, use the reinforcement learning algorithm to construct the excitation function and iteratively optimize to obtain the optimal control parameters.

2. The frequency-stable and energy-stable ignition method according to claim 1, characterized in that, The parameter prediction model includes a frequency feature network, a voltage feature network, an environmental feature network, and a feature fusion layer in parallel. The feature fusion layer is connected to the frequency feature network, the voltage feature network, and the environmental feature network, respectively. The frequency feature network, the voltage feature network, and the environmental feature network are used to extract feedback signals, discharge excitation signals, and environmental parameters, respectively.

3. The frequency-stable and energy-stable ignition method according to claim 2, characterized in that, The parameter prediction model also includes an attention layer. The front end of the feature fusion layer is provided with an attention layer, which is connected to the frequency feature network, voltage feature network, and environmental feature network, respectively.

4. The frequency-stable and energy-stable ignition method according to claim 2, characterized in that, Step S1 also includes preprocessing the data, filling in missing values, deleting outliers and duplicate values.

5. The frequency-stable and energy-stable ignition method according to claim 4, characterized in that, In step S1, the data is first completed, and then normalized.

6. A frequency-stable and energy-stable ignition method according to any one of claims 1-5, characterized in that, The environmental parameters include temperature and humidity.

Images