Micro-gas turbine combustion control method based on neural network

Through intelligent combustion control method based on neural networks, a multi-layer perceptron neural network and anomaly detection model is established to monitor and optimize the combustion process of micro-gas turbines in real time, solving the problem of prediction accuracy and slow calculation speed of combustion instability, and improving the Combustion stability and overall performance.

CN119084158AInactive Publication Date: 2024-12-06ZHEJIANG UNIV CITY COLLEGE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411551360.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-01
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prediction accuracy of existing micro-gas turbine combustion instability prediction models is not high, and the calculation speed is slow, so it cannot effectively suppress combustion instability, leading to noise and vibration problems, and may even cause catastrophic failures.

Method used

Using intelligent combustion control method based on neural network, design intelligent combustion controllers, establish a combustion instability prediction model for multi-layer perceptron neural network, combine anomaly detection model and optimization algorithm, monitor and optimize the combustion process in real time, and select through minimum value. The fuel flow is controlled by the device.

Benefits of technology

It improves combustion stability, overall control performance and reliability, enhances combustion efficiency and system stability, can quickly respond to abnormal situations and reduce the risk of failure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119084158A_ABST
    Figure CN119084158A_ABST
Patent Text Reader

Abstract

The invention provides a micro-gas turbine combustion control method based on a neural network, which is applied to a fuel flow control system and comprises the following specific steps: S1, designing an intelligent combustion controller based on the neural network, a starting speed increasing control signal, a rotating speed / power control signal, a power limiting control signal and an exhaust temperature control signal are obtained through the intelligent combustion controller based on the neural network; s2, establishing a combustion instability prediction model based on a neural network, training, and analyzing, evaluating and predicting performance; s3, performing real-time prediction by adopting a combustion instability prediction model; s4, the combustion process is monitored through the anomaly detection model, and corresponding diagnosis information is output when anomaly is detected; and S5, comparing a prediction result in the step S3 with a detection result in the step S4, judging the reasonability of combustion control, and performing optimization through an optimization algorithm.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of micro gas engine control and relates to a micro gas engine combustion control method based on a neural network. Background Art

[0002] A micro gas turbine is a miniaturized gas turbine, usually with a power of no more than 1000KW, and is mainly used in distributed power generation, drones, small power units, and some specific industrial applications. A micro gas turbine control system is a set of electronic components designed specifically to control and manage the operation of a micro gas turbine. The design and function of these control systems are key to ensuring efficient, stable, and safe operation of gas turbines under various operating conditions.

[0003] Combustion instability often occurs in the propulsion and power systems of micro-turbines. Combustion instability can cause high-amplitude noise and vibration, which may damage the structure of the combustion chamber or even cause catastrophic failures. Therefore, it must be suppressed and controlled. Generally, the indicators of combustion instability are predicted by establishing a combustion instability prediction model. The existing prediction models have low prediction accuracy and slow calculation speed, and thus cannot effectively suppress combustion instability. Summary of the invention

[0004] The present invention provides a micro-turbine combustion control method based on a neural network, which improves combustion stability and enhances overall control performance, combustion efficiency and reliability.

[0005] The technical solution adopted by the present invention is: A micro-turbine combustion control method based on a neural network is applied to a fuel flow control system, and the specific steps are as follows: S1, designing an intelligent combustion controller based on a neural network, and using the intelligent combustion controller based on the neural network to obtain a starting speed control signal, a speed / power control signal, a power limit control signal, and an exhaust temperature control signal; S2, establish a combustion instability prediction model based on a neural network, conduct training and analyze and evaluate the prediction performance; S3, real-time prediction using combustion instability prediction model; S4, using an anomaly detection model to monitor the combustion process and output corresponding diagnostic information when an anomaly is detected; S5, comparing the prediction result of step S3 with the detection result of step S4, judging the rationality of the combustion control, and optimizing it through an optimization algorithm.

[0006] Further, the control process of the neural network-based intelligent combustion controller in step S1 includes: obtaining the actual speed of the gas turbine, and obtaining a starting speed-up control signal based on the starting speed-up controller; obtaining the actual speed and actual power of the gas turbine, and obtaining a speed / power control signal based on the actual speed, actual power and corresponding given speed and given power of the gas turbine; obtaining a power limit control signal according to the given speed of the gas turbine; obtaining the actual exhaust temperature of the gas turbine, and obtaining an exhaust temperature control signal based on the corrected exhaust temperature and the exhaust temperature set value; Based on the starting speed control signal, speed / power control signal, power limit control signal and exhaust temperature control signal, a fuel flow control signal is obtained through a minimum selector, and the fuel flow control system is controlled based on the fuel flow control signal.

[0007] Furthermore, the combustion instability prediction model in step S2 is established as follows: The neural network adopts a multi-layer perceptron neural network, including an input layer, a hidden layer, and an output layer; The input data x of the input layer is various sensor data during the combustion process; The activation function of the hidden layer is γ, and the output h is ,in is the hidden layer weight matrix, is the bias vector of the hidden layer; The output layer is used to predict the index of combustion instability. The prediction y of the output layer is ,in is the output layer weight vector, is the bias of the output layer.

[0008] Further, the training and evaluation steps of the combustion instability prediction model in step S2 are as follows: Step S21, collecting historical data of various sensors during the combustion process and preprocessing them as a training data set; Step S22, training using the training data set, ; in are input features, is the corresponding label, m is the number of samples, and i is the i-th sample; Define the loss function of the neural network : ; in is the neural network for the sample The predicted output of is the corresponding true label; Step S23, use the back propagation algorithm to calculate the gradient of the loss function with respect to the parameter θ, and use the optimization algorithm to update the parameter : ; Where α is the learning rate, is the loss function The gradient with respect to the parameter θ; Step S24, using the validation set or the test set to evaluate the prediction performance of the neural network.

[0009] Further, the preprocessing in step S21 includes normalization processing, which is as follows: ; Where x is the input data, μ is the mean, is the standard deviation, X norm Represents normalized data.

[0010] Further, the evaluation index in step S24 includes the root mean square error: ; Where RMSE is the root mean square error.

[0011] Furthermore, the anomaly detection model in step S4 adopts a recurrent neural network, a long short-term memory network or a convolutional neural network.

[0012] Furthermore, when the anomaly detection model is trained, the neural network learns the patterns of the normal combustion process.

[0013] Furthermore, when the anomaly detection model is monitored, the neural network continuously evaluates the input data and detects instabilities based on the preset anomaly threshold or model prediction output.

[0014] Furthermore, the optimization algorithm in step S5 is a genetic algorithm, a network optimization, a fuzzy logic control algorithm, or a reinforcement learning algorithm.

[0015] Beneficial effects of the present invention: 1. Abnormal detection can improve combustion efficiency and reliability.

[0016] 2. After optimizing the control, the overall performance and stability can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of a fuel flow control system of the present invention; Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described below in conjunction with specific embodiments, but the present invention is not limited to these specific embodiments. Those skilled in the art should recognize that the present invention covers all possible alternatives, improvements and equivalents within the scope of the claims.

[0019] See also Figure 1 , Figure 2 The present invention provides a micro-turbine combustion control method based on a neural network, which is applied to a fuel flow control system. The specific steps are as follows: S1, designing an intelligent combustion controller based on a neural network, and using the intelligent combustion controller based on the neural network to obtain a starting speed control signal, a speed / power control signal, a power limit control signal, and an exhaust temperature control signal; The control process of the intelligent combustion controller based on the neural network includes: obtaining the actual speed of the gas turbine, and obtaining the starting speed control signal based on the starting speed controller; obtaining the actual speed and actual power of the gas turbine, and obtaining the speed / power control signal based on the actual speed, actual power and the corresponding given speed and given power of the gas turbine; obtaining the power limit control signal according to the given speed of the gas turbine; obtaining the actual exhaust temperature of the gas turbine, and obtaining the exhaust temperature control signal based on the corrected exhaust temperature and the exhaust temperature set value; Based on the starting speed control signal, the speed / power control signal, the power limit control signal and the exhaust temperature control signal, a fuel flow control signal is obtained through a minimum selector, and a fuel flow control system is controlled based on the fuel flow control signal; Specifically, the fuel servo control system of the gas turbine is as follows: Figure 1 As shown. During the actual operation of the gas turbine, the control system needs to control multiple parameters (speed, power or exhaust temperature). For each controlled parameter, an independent control loop will be formed, so that the fuel flow control system of the gas turbine contains multiple control loops that affect the fuel flow. The gas turbine control system generally includes loops such as the starting speed controller, the speed / power controller, the exhaust temperature controller and the power limit controller. Among them, the four control modules of the starting speed controller, the speed / power controller, the exhaust temperature controller and the power limit controller can all control the fuel flow of the gas turbine. During normal operation, each controller operates independently, and outputs the calculation results to the minimum selector for screening, and selects the minimum value to output to the fuel valve position controller to change the fuel flow. At this time, only one loop is in the control state, and the other loops are in the monitoring state, which can achieve non-disturbance switching between multiple loops.

[0020] S2, establish a combustion instability prediction model based on a neural network, conduct training and analyze and evaluate the prediction performance; the neural network identifies and predicts the characteristics and patterns of combustion instability by learning a large amount of experimental data or simulation results, and uses the neural network to predict indicators of combustion instability, such as the frequency or amplitude of the combustion wave.

[0021] The combustion instability prediction model is established as follows: The neural network adopts a multi-layer perceptron neural network, including an input layer, a hidden layer, and an output layer; The input data x of the input layer is various sensor data in the combustion process; the input data is such as the actual speed, actual power, exhaust temperature, etc. of the gas turbine. Suppose the input data is a vector , where n is the number of input features.

[0022] The hidden layer maps the input to a higher dimensional space through a series of linear transformations and nonlinear activation functions. The activation function of the hidden layer is γ, and the output h is ,in is the hidden layer weight matrix, is the bias vector of the hidden layer; The output layer has one neuron for predicting the index of combustion instability. The prediction y of the output layer is ,in is the output layer weight vector, is the bias of the output layer.

[0023] The training and evaluation steps of the combustion instability prediction model are as follows: Step S21, collect various sensor historical data during the combustion process, input data such as actual speed, actual power, exhaust temperature, etc. of the gas turbine, and perform preprocessing as a training data set; the preprocessing includes normalization processing to ensure the uniform scale of input features and improve the training speed and stability of the neural network, as follows: ; Where x is the input data, μ is the mean, is the standard deviation, X norm Represents normalized data.

[0024] Step S22, training using the training data set, ; in are input features, is the corresponding label, m is the number of samples, and i is the i-th sample; Define the loss function of the neural network : ; in is the neural network for the sample The predicted output of is the corresponding true label; Step S23, use the back propagation algorithm to calculate the gradient of the loss function with respect to the parameter θ, and use the optimization algorithm to update the parameter : ; Where α is the learning rate, is the loss function The gradient with respect to the parameter θ; Step S24, using the validation set or the test set to evaluate the prediction performance of the neural network.

[0025] The evaluation indicators include root mean square error RMSE: ; Where RMSE is the root mean square error.

[0026] S3, real-time prediction using combustion instability prediction model; S4, using an anomaly detection model to monitor the combustion process and output corresponding diagnostic information when an anomaly is detected; wherein the anomaly detection model uses a recurrent neural network, a long short-term memory network, or a convolutional neural network. When the anomaly detection model is trained, the neural network learns the pattern of the normal combustion process. When the anomaly detection model is monitored, the neural network continuously evaluates the input data and detects instability based on the preset anomaly threshold or the model prediction output.

[0027] Recurrent Neural Network (RNN) formula: For time step t, the hidden state h(t) and output y(t) are typically calculated as follows: ; ; in, is the input data, which can be various sensor data at a specific time point in the combustion system, W and b are weights and biases, and σ is the activation function, such as the sigmoid or tanh function.

[0028] Convolutional Neural Network (CNN) Formula: If CNN is used to process the spatial data of the combustion system (such as the temperature distribution in the combustion chamber), the following formula can be used: ; ; Here * represents the convolution operation, is the convolution kernel, is the bias term, is the activation function.

[0029] During the training phase, the neural network is back-propagated and optimized through a large amount of data on normal and abnormal situations to learn the characteristics of normal and abnormal patterns. Once the training is complete, the neural network can be used to monitor the operating status of the combustion system in real time.

[0030] When the neural network detects an anomaly, it can output corresponding diagnostic information, such as the specific anomaly type, possible causes, recommended repair measures, etc. This information can help operators or maintenance personnel respond and solve problems quickly, thereby improving the efficiency and reliability of the combustion system.

[0031] S5, comparing the prediction result of step S3 with the detection result of step S4, judging the rationality of the combustion control, and optimizing it through an optimization algorithm.

[0032] The optimization algorithm is a genetic algorithm, a network optimization, a fuzzy logic control algorithm or a reinforcement learning algorithm.

[0033] Genetic Algorithm: It optimizes control parameters by simulating natural selection and genetic mechanisms and is applicable to complex multivariable optimization problems. In combustion systems, genetic algorithms can be used to find the best combination of control parameters to improve the overall performance and stability of the system.

[0034] Network optimization: Use neural network models to model and predict the combustion system, and optimize the control strategy by optimizing the structure and parameters of the neural network. Use neural networks to predict combustion instability indicators, and adjust control parameters based on the prediction results to achieve intelligent optimization of the combustion system.

[0035] Fuzzy logic control algorithm: Fuzzy logic control is based on fuzzy set theory, can handle uncertainty and ambiguity, and is suitable for complex nonlinear system control. In the combustion system, fuzzy logic control can be used to design fuzzy reasoning systems, adjust control strategies according to real-time combustion data, and improve system stability and adaptability.

[0036] Reinforcement learning algorithm: Reinforcement learning learns how to take actions in a certain environment to maximize reward signals through the interaction between agents and the environment. It is applicable to long-term optimization and adaptive control problems. In the combustion system, reinforcement learning algorithms can be used to train intelligent agents to optimize control strategies based on real-time combustion data and environmental feedback to achieve the best system operation effect.

[0037] Select and combine according to the specific combustion system characteristics and optimization goals to achieve continuous improvement and optimization of system performance. Real-time monitoring and feedback control adjust the fuel flow to suppress the development of instability or maximize combustion efficiency.

[0038] The present invention can improve combustion efficiency and reliability through abnormality detection, and can improve overall performance and stability after optimized control.

Claims

1. A micro-turbine combustion control method based on neural network, applied to fuel flow control system, It is characterized in that The specific steps are as follows: S1, designing an intelligent combustion controller based on a neural network, and using the intelligent combustion controller based on the neural network to obtain a starting speed control signal, a speed / power control signal, a power limit control signal, and an exhaust temperature control signal; S2, establish a combustion instability prediction model based on a neural network, conduct training and analyze and evaluate the prediction performance; S3, real-time prediction using combustion instability prediction model; S4, using an anomaly detection model to monitor the combustion process and output corresponding diagnostic information when an anomaly is detected; S5, comparing the prediction result of step S3 with the detection result of step S4, judging the rationality of the combustion control, and optimizing it through an optimization algorithm.

2. According to the neural network-based micro-turbine combustion control method of claim 1, Features: The control process of the neural network-based intelligent combustion controller in step S1 includes: obtaining the actual speed of the gas turbine, and obtaining a starting speed-up control signal based on the starting speed-up controller; obtaining the actual speed and actual power of the gas turbine, and obtaining a speed / power control signal based on the actual speed, actual power and corresponding given speed and given power of the gas turbine; obtaining a power limit control signal according to the given speed of the gas turbine; obtaining the actual exhaust temperature of the gas turbine, and obtaining an exhaust temperature control signal based on the corrected exhaust temperature and the exhaust temperature set value; Based on the starting speed control signal, speed / power control signal, power limit control signal and exhaust temperature control signal, a fuel flow control signal is obtained through a minimum selector, and the fuel flow control system is controlled based on the fuel flow control signal.

3. According to the neural network-based micro-turbine combustion control method of claim 1, Features: The combustion instability prediction model in step S2 is established as follows: The neural network adopts a multi-layer perceptron neural network, including an input layer, a hidden layer, and an output layer; The input data x of the input layer is various sensor data during the combustion process; The activation function of the hidden layer is γ, and the output h is ,in is the hidden layer weight matrix, is the bias vector of the hidden layer; The output layer is used to predict the index of combustion instability. The prediction y of the output layer is ,in is the output layer weight vector, is the bias of the output layer.

4. A micro-turbine combustion control method based on neural network according to claim 1, Features: The training and evaluation steps of the combustion instability prediction model in step S2 are as follows: Step S21, collecting historical data of various sensors during the combustion process and preprocessing them as a training data set; Step S22, training using the training data set, ; in are input features, is the corresponding label, m is the number of samples, and i is the i-th sample; Define the loss function of the neural network : ; in is the neural network for the sample The predicted output of is the corresponding true label; Step S23, use the back propagation algorithm to calculate the gradient of the loss function with respect to the parameter θ, and use the optimization algorithm to update the parameter : ; Where α is the learning rate, is the loss function The gradient with respect to the parameter θ; Step S24, using the validation set or the test set to evaluate the prediction performance of the neural network.

5. A micro-turbine combustion control method based on neural network according to claim 4, Features: The preprocessing in step S21 includes normalization processing, which is as follows: ; Where x is the input data, μ is the mean, is the standard deviation, X norm Represents normalized data.

6. A micro-turbine combustion control method based on neural network according to claim 5, Features: The evaluation indicators in step S24 include the root mean square error: ; Where RMSE is the root mean square error.

7. A micro-turbine combustion control method based on neural network according to claim 1, Features: In step S4, the anomaly detection model adopts a recurrent neural network, a long short-term memory network or a convolutional neural network.

8. A micro-turbine combustion control method based on neural network according to claim 7, Features: When the anomaly detection model is trained, the neural network learns the patterns of the normal combustion process.

9. A micro-turbine combustion control method based on neural network according to claim 8, Features: When monitoring anomaly detection models, the neural network continuously evaluates the input data and detects instabilities based on preset anomaly thresholds or model prediction outputs.

10. A micro-turbine combustion control method based on neural network according to claim 1, Features: The optimization algorithm in step S5 is a genetic algorithm, a network optimization, a fuzzy logic control algorithm, or a reinforcement learning algorithm.