CN118622478B - Gas turbine OTC temperature optimization method and system based on MBR thermocouple - Google Patents

Abstract

The present application provides a gas turbine OTC temperature optimization method and system based on MBR thermocouples, which relates to the technical field of data mining and analysis. The method includes: when the gas turbine is in the no-load free rotation stage, the real-time OTC temperature is determined based on the temperature signal matrix collected by the first thermocouple array; the real-time temperature mean and the real-time temperature variance are calculated according to the real-time OTC temperature, and the temperature change rate mean is calculated according to the real-time OTC temperature and the historical OTC temperature group corresponding to the adjacent preset time period, and the OTC characteristic variables are determined according to the real-time temperature mean, the real-time temperature variance and the temperature change rate mean; the OTC characteristic variables are input into the parameter optimization model to determine the optimized control parameters for the gas turbine, and the parameter optimization model adopts the reinforcement learning model. Thus, the OTC temperature changes are monitored and mined and analyzed in real time based on reinforcement learning, so as to realize the adaptive optimization of the system control parameters.

Description

Gas turbine OTC temperature optimization method and system based on MBR thermocouple

Technical Field

The present application relates to the technical field of data mining and analysis, and in particular to a gas turbine OTC temperature optimization method and system based on MBR thermocouples.

Background Art

A gas turbine is a thermal engine that converts the heat released by fuel combustion into mechanical energy. Its efficient energy conversion capability makes it an important equipment in modern industry and transportation. Gas turbines mainly use temperature monitoring to implement OTC (Outlet Temperature Control) operation to ensure efficient, stable and safe operation.

At present, the temperature monitoring of gas turbines mainly relies on thermocouple sensors. Thermocouple sensors are temperature sensors that use the temperature difference at the contact point of two different materials to generate electromotive force. They have a simple structure, fast response, and a wide measurement range. Most gas turbine temperature monitoring systems rely on traditional temperature sensors, including common K-type, J-type and other ordinary thermocouples, which are often arranged between the combustion chamber and the turbine. When the turbine runs at high speed, electromagnetic noise is often frequent, which can easily cause measurement errors.

Traditional OTC systems usually rely on preset static control parameters, but the operating environment and working conditions of gas turbines are complex and changeable, and static control parameters cannot respond to these changes in time, resulting in fluctuations and instability in combustion efficiency. With the development of information technology, computer-aided technology and some basic data analysis methods have gradually been introduced into the control system of gas turbines, such as the A-SMC system (Advanced Combustion Margin Control System for Gas Turbines), but it still needs to be improved in data analysis and intelligent decision-making.

In response to the above problems, the industry has not yet proposed a better technical solution.

Summary of the invention

The present application provides a gas turbine OTC temperature optimization method, system, storage medium, computer program product and electronic device based on MBR thermocouple, which is used to at least solve the problem that the static control method of the gas turbine based on sensor measurement of temperature in the current related technology cannot meet the needs of dynamic intelligent decision-making, resulting in unstable system control effect.

In a first aspect, an embodiment of the present application provides a gas turbine OTC temperature optimization method based on MBR thermocouples, comprising: when the gas turbine is in a no-load free rotation stage, determining the real-time OTC temperature based on a temperature signal matrix collected by a first thermocouple array; the first thermocouple array comprises a plurality of first thermocouples distributed in a first area adjacent to an exhaust port, the first thermocouples adopt band response thermocouples, and each of the first thermocouples has a unique corresponding measurement wavelength; calculating the real-time temperature mean and the real-time temperature variance according to the real-time OTC temperature, and calculating the temperature change rate mean according to the real-time OTC temperature and the historical OTC temperature group corresponding to the adjacent preset time period, and calculating the temperature change rate mean according to the real-time OTC temperature and the historical OTC temperature group corresponding to the adjacent preset time period. The real-time temperature mean, the real-time temperature variance and the temperature change rate mean determine the OTC characteristic variable; the OTC characteristic variable is input into the parameter optimization model to determine the optimized control parameters for the gas turbine; the optimized control parameters include at least one of the following: fuel supply, air supply, inlet guide vane angle and exhaust temperature setting value; the parameter optimization model adopts a reinforcement learning model; the state of the reinforcement learning model is defined by the real-time OTC temperature characteristic variable; the action of the reinforcement learning model is defined by the adjustment information of the gas turbine control parameters; the reward of the reinforcement learning model is defined by the combustion efficiency and pollutant emission level of the gas turbine.

In a second aspect, an embodiment of the present application provides a gas turbine OTC temperature optimization system based on MBR thermocouples, comprising: a temperature acquisition unit, used to determine the real-time OTC temperature based on the temperature signal matrix collected by the first thermocouple array when the gas turbine is in the no-load free rotation stage; the first thermocouple array includes a plurality of first thermocouples distributed in a first area adjacent to the exhaust port, the first thermocouples adopt band response thermocouples, and each of the first thermocouples has a unique corresponding measurement wavelength; a variable extraction unit, used to calculate the real-time temperature mean and the real-time temperature variance according to the real-time OTC temperature, and calculate the temperature change rate mean according to the real-time OTC temperature and the historical OTC temperature group corresponding to the adjacent preset time period, The OTC characteristic variable is determined according to the real-time temperature mean, the real-time temperature variance and the temperature change rate mean; an optimization parameter determination unit is used to input the OTC characteristic variable into a parameter optimization model to determine the optimized control parameters for the gas turbine; the optimized control parameters include at least one of the following: fuel supply amount, air supply amount, inlet guide vane angle and exhaust temperature setting value; the parameter optimization model adopts a reinforcement learning model; the state of the reinforcement learning model is defined by the real-time OTC temperature characteristic variable; the action of the reinforcement learning model is defined by the adjustment information of the gas turbine control parameters; the reward of the reinforcement learning model is defined by the combustion efficiency and pollutant emission level of the gas turbine.

According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the gas turbine OTC temperature optimization method based on MBR thermocouples according to any embodiment of the present application.

In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the gas turbine OTC temperature optimization method based on MBR thermocouples of any embodiment of the present application are implemented.

In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program/instruction, which, when executed by a processor, implements the steps of the gas turbine OTC temperature optimization method based on MBR thermocouples of any embodiment of the present application.

The gas turbine OTC temperature optimization method based on MBR thermocouple provided in this application can produce at least the following technical effects:

(1) By calculating the temperature mean, temperature variance and temperature change rate mean in real time, the OTC feature variables are dynamically monitored and updated. The state of the reinforcement learning model is defined by the real-time OTC feature variables, and the action is defined according to the adjustment information of the gas turbine control parameters. By using the combustion efficiency and pollutant emission level as the reward function, the reinforcement learning model can continuously optimize the control strategy, so that the system can adaptively learn and adjust the control parameters of the gas turbine, and can dynamically adjust the control parameters according to these real-time feature variables. The gas turbine can maintain the optimal operating state under different operating conditions, thereby enhancing the dynamic response capability of the system.

(2) In terms of signal sampling, compared with installing thermocouples between the combustion chamber and the turbine or in the area at the combustion chamber outlet, the present technical solution uses the first thermocouple array near the exhaust port for sampling, which can provide more uniform temperature data, and use it to calculate the OTC temperature in the stable operation stage after the anti-surge valve is closed, effectively reducing the impact of electromagnetic interference noise generated by the high-speed rotation of the turbine on the temperature sensing detection parameters. In addition, by arranging the MBR (Multi-Band Response) thermocouple array, each band response thermocouple can capture infrared radiation signals of different wavelengths, reducing the deviation caused by single wavelength measurement, thereby providing multi-spectrum temperature measurement data, which can more accurately capture subtle changes in exhaust temperature.

Through this technical solution, OTC temperature changes can be monitored and analyzed in real time based on reinforcement learning, trends and changes can be captured in a timely manner, and control parameters can be adaptively adjusted under different operating conditions, thereby achieving adaptive optimization of system control parameters and enhancing the intelligent analysis and decision-making capabilities of the gas turbine control system.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

FIG1 shows a flow chart of an example of a gas turbine OTC temperature optimization method based on an MBR thermocouple according to an embodiment of the present application;

FIG2 shows a schematic structural connection diagram of an example of a gas turbine suitable for applying the gas turbine OTC temperature calculation method of the embodiment of the present application;

FIG3 is a schematic diagram showing the working principle of an example of a state transfer action in a reinforcement learning model;

FIG4 shows a schematic diagram of a structural connection of an example of a parameter optimization model according to an embodiment of the present application;

FIG5 shows a structural block diagram of an example of a gas turbine OTC temperature optimization system based on MBR thermocouples according to an embodiment of the present application;

FIG. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

FIG1 shows a flow chart of an example of a gas turbine OTC temperature optimization method based on MBR thermocouples according to an embodiment of the present application.

Regarding the executor of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities. By adopting MBR thermocouples, real-time OTC characteristic variable calculation and reinforcement learning model, the temperature monitoring accuracy, dynamic response capability and control parameter optimization capability of the gas turbine OTC system are significantly improved, thereby improving the combustion efficiency of the gas turbine, reducing pollutant emissions, and enhancing the stability and safety of the system.

In some examples, it can be an OTC temperature optimization platform, and can be integrated and configured in an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of terminal or electronic device can be diverse, such as a mobile phone, tablet computer, or desktop computer, etc.

As shown in FIG. 1 , in step S110 , when the gas turbine is in the no-load free rotation stage, the real-time OTC temperature is determined based on the temperature signal matrix collected by the first thermocouple array.

In some embodiments, the FSNL (Full Speed No Load, no-load free rotation) stage can be achieved by detecting a specific signal. More specifically, the anti-surge valve closing signal can be monitored to identify whether the gas turbine is in the no-load free rotation stage. Here, during the operation of the gas turbine, the anti-surge valve is used to prevent the gas turbine from surging under specific operating conditions. The status signal of the anti-surge valve can be obtained by a sensor installed on the valve. When the valve is closed, the sensor outputs a corresponding closed state signal. By detecting the status signal of the anti-surge valve, the working state of the gas turbine, such as a stable operating state or other state, can be accurately judged. In addition, when the valve is opened, the sensor outputs a corresponding opening signal, and the gas turbine is in the startup stage or low-load condition.

FIG2 shows a schematic diagram of the structural connection of an example of a gas turbine suitable for applying the gas turbine OTC temperature calculation method of the embodiment of the present application. As shown in FIG2, according to the direction of the airflow, the gas turbine includes a compressor, a combustion chamber, a turbine and a diffuser, and finally discharges the gas through the exhaust port, for example, the exhaust gas will flow to the boiler, etc. Here, the compressor is responsible for compressing the air and sending it into the combustion chamber, and the air is mixed and burned with the fuel in the combustion chamber. The high-temperature and high-pressure gas generated by the combustion drives the turbine to rotate to output mechanical energy, and the diffuser can decelerate and expand the flowing gas to reduce the airflow speed. It should be noted that in some current non-public or potential research technologies, a thermocouple is set between the combustion chamber and the turbine, for example, by setting a thermocouple in the second region plane to sample the combustion chamber exhaust temperature as the OTC temperature, but in the high-speed operation stage of the turbine, for example, the FSNL stage, the electromagnetic interference generated by the turbine will affect the temperature sampling accuracy.

According to the embodiment of the present application, a plurality of first thermocouples are arranged in the first area adjacent to the exhaust port to form a first thermocouple array, each of which is a band-responsive thermocouple, and each of which has a unique corresponding measurement wavelength, thereby arranging an MBR thermocouple array in the plane of the first area. Here, the thermocouple can respond sensitively to temperature changes within a specific band, thereby reducing the measurement error caused by electromagnetic noise.

It should be noted that when the anti-surge valve is in the closed state, the gas turbine has reached a stable operating state from startup, for example, in the FSNL stage, calling the MBR thermocouple array sampling temperature information of the first area plane can effectively reduce the electromagnetic noise level, and the plane temperature part will be more uniform, achieving high-quality temperature sampling information. Furthermore, by using the MBR thermocouple array, the exhaust temperature change information at multiple wavelengths can be obtained at the same time, realizing the refined measurement of the temperature field of the entire first area plane.

In one example of the embodiment of the present application, the temperature signal matrix collected by the first thermocouple array can be used as the real-time OTC temperature. In another example of the embodiment of the present application, the OTC temperature optimization platform can also calibrate and optimize the temperature signal matrix collected by the sensor to obtain the real-time OTC temperature.

In step S120, the real-time temperature mean and the real-time temperature variance are calculated according to the real-time OTC temperature, and the temperature change rate mean is calculated according to the real-time OTC temperature and the historical OTC temperature group corresponding to the adjacent preset time period, and the OTC characteristic variable is determined according to the real-time temperature mean, the real-time temperature variance and the temperature change rate mean.

In some embodiments, by calculating the average temperature value of all thermocouple positions in the temperature signal matrix, the real-time temperature mean is determined, which can provide the overall temperature level of the entire plane and eliminate the influence of local temperature fluctuations. By calculating the variance of the temperature values in the temperature signal matrix, the real-time temperature variance is determined, which can measure the uniformity of the temperature distribution and help identify the fluctuation of the temperature field. In addition, by calculating the average value of the temperature change rate of all thermocouple positions through the real-time OTC temperature and the historical OTC temperature group, it can provide the overall temperature change trend and help predict the dynamic change of the temperature field.

More specifically, the real-time temperature average is calculated in the following way:

, Formula (1)

In the formula, represents the number of thermocouples in the first thermocouple array, Indicates The real-time temperature value of the thermocouple, Indicates the real-time average temperature.

The real-time temperature variance is calculated as follows:

, Formula (2)

In the formula, Represents the real-time temperature variance, which is used to measure the uniformity of the real-time temperature distribution.

The mean temperature change rate is calculated as follows:

, Formula (3)

In the formula, is the mean temperature change rate, which indicates the average rate at which temperature changes over time; Indicates The temperature value of a thermocouple at the last sampling time point; Indicates the length of the preset time period.

Through the data processing method of this embodiment, the temperature mean is used to provide the overall temperature level of the entire first regional plane, which is the basis for judging the current temperature state. The temperature variance measures the uniformity of the temperature distribution, and can identify the fluctuations and non-uniformities in the temperature field. The mean of the temperature change rate reflects the trend and rate of temperature change, which helps to predict the dynamic behavior of temperature change and facilitates real-time adjustment and control. Thus, a comprehensive understanding of the current operating state can be provided.

In one example of the embodiment of the present application, the real-time temperature mean, real-time temperature variance and temperature change rate mean can be directly determined as OTC feature variables. In another example of the embodiment of the present application, further analysis and processing can be performed to enrich the information entropy of the OTC feature variables.

More specifically, the temperature mean, real-time temperature variance, and temperature change rate mean are further processed using the standardization and weighted fusion method to determine the OTC feature variables:

, Formula (4)

, Formula (5)

, Formula (6)

, Formula (7)

In the formula, , and Respectively represent the standardized real-time temperature mean, real-time temperature variance and mean of temperature change rate; represents the historical mean of the mean temperature, represents the historical standard deviation of the mean temperature, represents the historical mean of temperature variance, represents the historical standard deviation of the temperature variance, represents the historical mean of the mean temperature change rate, The historical standard deviation of the mean temperature change rate is calculated based on the historical records of the preset window period; represents the OTC characteristic variable, They represent the parameter weight coefficients of the corresponding variable parameters respectively.

The dimensional differences between different characteristic variables are eliminated through standardization, so that these characteristic variables have a unified standard when calculated and analyzed, which can more accurately reflect the state and changes of the system and avoid analysis errors caused by data scale differences. In addition, the use of weighted fusion method to integrate multiple characteristic variables can more accurately extract the core information of the gas turbine operating status. In addition, by reasonably setting the weight coefficient, the importance of different characteristic variables in describing the system status can be highlighted.

Parameter weight coefficients for each variable parameter , in the sense of the embodiment of the present application, it can be a preset fixed value. In another example of the embodiment of the present application, It is adaptively adjusted through initial value setting and based on the fuzzy rule base and real-time gas turbine performance indicator feedback.

Here, the fuzzy rule base includes a plurality of fuzzy rules, and each fuzzy rule includes a fuzzy state of a calibration variable parameter, a calibration engine performance indicator feedback, and a calibration adjustment amount for a corresponding parameter weight coefficient.

Specifically, first, define fuzzy sets. Specifically, according to different ranges of characteristic variables (such as temperature variance, mean temperature change rate), define corresponding fuzzy sets. Each fuzzy set represents a fuzzy state of the characteristic variable, for example, temperature variance: {low, medium, high}, mean temperature change rate: {stable, fluctuating}, and so on. Then, define fuzzy rules, and define a fuzzy rule base based on practical experience and expert knowledge. Each rule associates the fuzzy state of the input characteristic variable with the suggestion of output weight adjustment. For example, if "temperature variance" is "high" and "system stability" is "decreasing", then "increase the weight of temperature variance"; if "mean temperature change rate" is "stable" and "combustion efficiency" is "improved", then "reduce the weight of mean temperature change rate", and so on. Then, according to the fuzzy rule base and the fuzzy set of the current characteristic variable, calculate the activation degree of each rule, which can be calculated according to the input membership degree, which reflects the degree to which the rule is satisfied. According to the activated rules and their corresponding actions (such as increasing or decreasing the weight coefficient), calculate the result of the fuzzy output. Furthermore, the fuzzy output is converted into an accurate adjustment value through the centroid method, that is, the accurate output value is obtained by calculating the centroid of the fuzzy output.

Through the embodiment of the present application, the system can sense the operating status of the gas turbine in real time by comprehensively utilizing the fuzzy rule base and system feedback, and adjust the weight coefficient accordingly. Thus, by introducing fuzzy logic control, the system can show stronger robustness when facing complex and uncertain operating environments.

In step S130, the OTC characteristic variables are input into a parameter optimization model to determine the optimized control parameters for the gas turbine, wherein the optimized control parameters include at least one of the following: fuel supply, air supply, inlet guide vane angle, and exhaust temperature setting value, and the parameter optimization model adopts a reinforcement learning model.

FIG3 is a schematic diagram showing the working principle of an example of a state transfer action in a reinforcement learning model.

As shown in Figure 3, the state transition diagram involves multiple basic states. ~ The state space composed of may have state transitions between different basic states, such as Indicates from arrive The state transition, Indicates from arrive The state transition, Indicates from arrive Here, the corresponding state transition can occur based on the state transition strategy, and each state transition strategy can be used to cause different state transitions. The state transition strategy can cause state transition or .

In addition, the range of states to which a base state in the state space can be transferred (also called transferable states) is generally restricted or conditional, e.g. ~ Neither of them will ~ There is a state transition between The states that can be transferred to are and ,etc.

In some examples of the embodiments of the present application, the state of the reinforcement learning model is defined by the real-time OTC temperature characteristic variable, the action of the reinforcement learning model is defined by the adjustment information of the gas turbine control parameters, and the reward of the reinforcement learning model is defined by the combustion efficiency and pollutant emission level of the gas turbine. It should be understood that the environment refers to the operating system of the gas turbine, including its physical state and control mechanism. In the reinforcement learning model, the environment will return the corresponding state and reward for each action. The action is the control operation that the agent can perform in the current state, by adjusting the fuel supply to change the fuel flow rate, by adjusting the air supply to adjust the amount of air entering the combustion chamber, by adjusting the inlet guide vane (IGV, Inlet Guide Vane) angle to change the opening of the inlet guide vane, updating the exhaust temperature set value, and so on.

Through the embodiments of the present application, the reinforcement learning model can find the target action or optimal control strategy corresponding to the maximum reward in a variable state space or operating environment through continuous learning and adjustment, and can adaptively adjust the control parameters under different operating conditions, thereby improving the combustion efficiency of the gas turbine and reducing pollutant emissions.

In some examples of the embodiments of the present application, the first thermocouple array (or, MBR thermocouple array) includes a first band response thermocouple group and a second band response thermocouple group, and each thermocouple in the first band response thermocouple group is uniformly distributed in the flue gas core flow area according to a first distribution density, and each thermocouple in the second band response thermocouple group is uniformly distributed in other areas of the first area except the flue gas core flow area according to a second distribution density, wherein the first distribution density is greater than the second distribution density.

It should be noted that the flue gas core flow area is the main flow area of the boiler flue gas, which is usually located in the middle and upper part of the first area plane. Since this area is the main flow of high-temperature flue gas, the temperature changes rapidly and drastically. Therefore, designing a denser thermocouple layout for this sub-area can be more helpful in monitoring the temperature changes in this area, helping to understand the state of the combustion process and the completeness of fuel combustion. For other areas, such as flow separation areas and cooling areas, thermocouples can be deployed in small quantities to save sensor system resources.

Through this embodiment, a dense first-band response thermocouple group is arranged in the flue gas core flow area, which significantly improves the temperature monitoring accuracy of the area, can capture the subtle temperature changes and gradients in the high-temperature flue gas flow path, and help to fully understand the dynamic behavior of the combustion process. By arranging the second-band response thermocouple group in other areas, although the density is low, it can still ensure that the temperature field of the entire first area plane is fully covered, realizing global monitoring of the temperature field.

In some examples of the embodiments of the present application, the parameter optimization model adopts a deep Q- network (Deep Q-Network, DQN). FIG4 shows a schematic diagram of a structural connection of an example of a parameter optimization model according to an embodiment of the present application.

As shown in FIG. 4 , the parameter optimization model 400 includes an input layer 410 , a hidden layer 420 , and an output layer 430 .

More specifically, the deep Q network is used to determine the optimal control parameters by performing the following operations: The input layer 410 is used to receive the state vector corresponding to the OTC feature variable.

The hidden layer 420 is used to process the state vector through a multi-layer neural network to extract at least one state-action mapping relationship, thereby obtaining a corresponding plurality of potential actions.

In some embodiments, the hidden layer 420 may use a multi-layer neural network to extract features and learn complex state-action mapping relationships. For example, a 2-3 layer neural network may be configured, each layer having 128-256 neurons, and using a ReLU activation function.

The output layer 430 is used to determine the Q value corresponding to each potential action, and determine the optimized control parameter according to the adjustment information of the gas turbine control parameter corresponding to the potential action with the maximum Q value. It should be noted that the Q value is defined by the expected cumulative reward that can be obtained by taking the corresponding potential action under the state vector. In some embodiments, for the discrete action space corresponding to each potential action, the Q value of each potential action is output, and then the potential action with the maximum Q value is selected to determine the optimized control parameter, so as to realize the dynamic optimization decision of the system.

During the training process of the deep Q network, an experience replay mechanism is used to randomly extract small batches of data from past experience for training, which helps to break data correlation and stabilize the network training process.

In addition, the target Q network can be introduced To calculate the target Q value, in order to reduce the estimated deviation, and the parameters of the target Q network Regular updates are made from the main network (i.e., the deep Q network), for example, the parameters are copied and updated from the deep Q network every certain number of training steps to achieve online learning. Furthermore, the epsilon-greedy strategy can be used to balance exploration and exploitation. As training progresses, the epsilon value is gradually reduced to reduce random exploration, and in the later stages of training, the algorithm can more stably utilize the best strategy actions that have been learned to ensure higher rewards.

Through the embodiment of the present application, the parameter optimization model using the deep Q network has adaptive learning capabilities and can optimize the control parameters of the gas turbine in real time in a constantly changing environment. Through online learning and continuously updated Q value functions, the model can quickly adapt to different operating conditions and environmental conditions to achieve adaptive optimization control.

As a further optimization, the deep Q network is trained based on a Bayesian Neural Network (BNN) through a data sample set. Through the embodiments of the present application, the Bayesian Neural Network is used to optimize the training of the deep Q network, which can model the probability distribution of the Q value, handle uncertainty, and optimize the action selection strategy.

In this embodiment, the Bayesian neural network quantifies uncertainty by introducing a probability distribution for the weights of the network instead of fixed weights. The BNN is used to estimate the Q value of the state-action pair and provide a probability distribution for each Q value. More specifically, assuming that the Q value follows a normal distribution, the Bayesian neural network is used to output the mean and variance of the Q value for a given state and action corresponding to the sample:

, Formula (8)

In the formula, Indicates the status Action Q value; Represents a mean and variance Normal distribution of represents the state predicted by the Bayesian neural network and actions The corresponding mean of the Q value distribution, represents the parameters of the Bayesian neural network; Represents the state predicted by the Bayesian neural network and actions The variance of the corresponding Q- value distribution.

During the training process, the agent interacts with the environment to generate data samples , and stored in the experience replay pool. Through this embodiment, BNN can quantify the uncertainty of the prediction by introducing probability distribution for Q value, which helps the system to identify and manage risks in the decision-making process. For example, when the model faces a state with high uncertainty, it can adopt a conservative strategy to reduce potential risks. By modeling the mean and variance of the Q value distribution, in scenarios involving multi-objective optimization (such as combustion efficiency and pollutant emissions), BNN can better balance and optimize multiple objectives and provide more accurate control parameter optimization suggestions.

In some embodiments, the parameters of the BNN are updated by using a back-propagation algorithm and a gradient descent method. , to minimize the loss function. In addition, the parameters of the target Q network are Regularly updated as the parameters of BNN As training progresses, increasing the diversity and coverage of the data sample set, BNN is able to continuously update the estimated distribution of Q values to gradually reduce uncertainty.

In some examples of the embodiments of the present application, the loss function of the Bayesian neural network considers not only the expected value of the Q value, but also its uncertainty. Here, the negative log-likelihood function (NLL) is used to measure the difference between the Q value distribution output by the model and the actual target Q value. In addition, uncertainty regularization loss is introduced in the loss function to avoid overconfidence of the model.

More specifically, the loss function of the Bayesian neural network is:

, formula (9)

, Formula (10)

In the formula, Represents the parameters of the Bayesian neural network The corresponding total loss is Represents the total number of samples in the data sample set; represents the negative log-likelihood loss term, Represents the target Q value of the i- th sample The probability density under the normal distribution predicted by the Bayesian neural network is the mean corresponding to the i- th sample. and variance describe; represents the target Q value corresponding to the i - th sample, represents the mean of the Q value distribution predicted by the Bayesian neural network for the i -th sample, represents the variance of the Q value distribution predicted by the Bayesian neural network for the i- th sample; represents the uncertainty regularization term; Represents the regularization coefficient, which is used to control the weight of the regularization term; Variance Weighted by the reciprocal of ; is the immediate reward of the i- th sample, and is the corresponding state Next select action Rewards received after is a discount factor that measures the importance of future rewards; is the Boolean flag of the i- th sample; Indicates that the current round of the i- th sample has ended, and there is no subsequent state transfer action. At this time, the target Q value is only the immediate reward ; Indicates that the current round has not ended, there are subsequent state transfer actions and rewards, and the target Q value is the current reward Plus the maximum expected cumulative reward The discount value of Indicates execution of an action After reaching the new state, In the new state The potential action below, represents the new state estimated by the target Q network and actions The corresponding Q value is, are the parameters of the target Q network.

Through this embodiment, using the negative log-likelihood term in the loss function, BNN can accurately fit the target Q value distribution, which helps the intelligent agent to select the optimal control action under different states. Using the uncertainty regularization term in the loss function, the model can effectively manage the uncertainty of the predicted Q value distribution. Specifically, by adjusting the variance, the model can identify areas of high uncertainty, thereby taking more cautious strategies and avoiding high-risk decisions. In addition, through the introduction of the regularization term, the loss function can effectively avoid the overfitting of the model to the training data, especially when facing high-dimensional feature space, BNN can maintain good generalization ability of the data.

It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

FIG5 shows a structural block diagram of an example of a gas turbine OTC temperature optimization system based on MBR thermocouples according to an embodiment of the present application.

As shown in FIG. 5 , the gas turbine OTC temperature optimization system 500 based on MBR thermocouples includes a temperature acquisition unit 510 , a variable extraction unit 520 and an optimization parameter determination unit 530 .

The temperature acquisition unit 510 is used to determine the real-time OTC temperature based on the temperature signal matrix collected by the first thermocouple array when the gas turbine is in the no-load free rotation stage; the first thermocouple array includes a plurality of first thermocouples distributed in a first area adjacent to the exhaust port, the first thermocouples are band response thermocouples, and each of the first thermocouples has a unique corresponding measurement wavelength.

The variable extraction unit 520 is used to calculate the real-time temperature mean and the real-time temperature variance according to the real-time OTC temperature, and calculate the temperature change rate mean according to the real-time OTC temperature and the historical OTC temperature group corresponding to the adjacent preset time period, and determine the OTC characteristic variable according to the real-time temperature mean, the real-time temperature variance and the temperature change rate mean.

The optimization parameter determination unit 530 is used to input the OTC characteristic variables into a parameter optimization model to determine the optimization control parameters for the gas turbine; the optimization control parameters include at least one of the following: fuel supply amount, air supply amount, inlet guide vane angle and exhaust temperature setting value.

The parameter optimization model adopts a reinforcement learning model; the state of the reinforcement learning model is defined by the real-time OTC temperature characteristic variable; the action of the reinforcement learning model is defined by the adjustment information of the gas turbine control parameters; the reward of the reinforcement learning model is defined by the combustion efficiency and pollutant emission level of the gas turbine.

In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of any of the above-mentioned gas turbine OTC temperature optimization methods based on MBR thermocouples in the present application.

In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes any step of the above-mentioned gas turbine OTC temperature optimization method based on MBR thermocouples.

In some embodiments, the embodiments of the present application also provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the gas turbine OTC temperature optimization method based on MBR thermocouples.

FIG6 is a schematic diagram of the hardware structure of an electronic device for executing a gas turbine OTC temperature optimization method based on an MBR thermocouple according to another embodiment of the present application. As shown in FIG6 , the device includes:

One or more processors 610 and a memory 620 , with one processor 610 being taken as an example in FIG. 6 .

The device for executing the gas turbine OTC temperature optimization method based on MBR thermocouple may further include: an input device 630 and an output device 640 .

The processor 610, the memory 620, the input device 630 and the output device 640 may be connected via a bus or other means, and FIG6 takes the connection via a bus as an example.

The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions/modules corresponding to the gas turbine OTC temperature optimization method based on MBR thermocouples in the embodiment of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 620, that is, the gas turbine OTC temperature optimization method based on MBR thermocouples in the above method embodiment is realized.

The memory 620 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 620 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include a memory remotely arranged relative to the processor 610, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The input device 630 can receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 640 can include a display device such as a display screen.

The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, the gas turbine OTC temperature optimization method based on MBR thermocouple in any of the above method embodiments is executed.

The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

(1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

(2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

(3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

(4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM/RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (5)

1. An MBR thermocouple-based gas turbine OTC temperature optimization method, comprising:

Determining a real-time OTC temperature based on a temperature signal matrix acquired by the first thermocouple array when the gas turbine is in an idle free rotation stage; the first thermocouple array comprises a plurality of first thermocouples distributed in a first area adjacent to the exhaust port, the first thermocouples adopt wave band response thermocouples, and each first thermocouple respectively has a unique corresponding measurement wavelength;

Calculating a real-time temperature mean value and a real-time temperature variance according to the real-time OTC temperature, calculating a temperature change rate mean value according to the real-time OTC temperature and a historical OTC temperature group corresponding to a preset time period, and determining an OTC characteristic variable according to the real-time temperature mean value, the real-time temperature variance and the temperature change rate mean value;

Inputting the OTC feature variables to a parameter optimization model to determine optimized control parameters for the gas turbine; the optimal control parameters include at least one of: fuel supply, air supply, inlet guide vane angle, and exhaust temperature setpoint;

the parameter optimization model adopts a deep Q network, wherein the deep Q network comprises an input layer, a hidden layer and an output layer;

the deep Q network is configured to determine the optimal control parameters by performing the following operations;

the input layer is used for receiving a state vector corresponding to the OTC characteristic variable;

the hidden layer is used for processing the state vector through a multi-layer neural network to extract at least one state-action mapping relation so as to obtain a plurality of corresponding potential actions;

The output layer is used for determining the Q value corresponding to each potential action and determining the optimized control parameter according to the adjustment information of the gas turbine control parameter corresponding to the potential action with the maximum Q value; wherein the Q value is defined by the expected jackpot that would be achieved by taking the corresponding potential action under the state vector;

the deep Q network is trained through a set of data samples based on a bayesian neural network;

the bayesian neural network is used for outputting the mean value and the variance of the Q value for a given state and action corresponding to the sample:

，

In the method, in the process of the invention, Representation is directed to stateAction of (2)Q value of (2); The representation has an average value Sum of variancesIs a normal distribution of (2); representing states predicted by a bayesian neural network And actionsThe corresponding mean value of the Q-value distribution,Parameters representing a bayesian neural network; representing states predicted by a Bayesian neural network And actionsVariance of the corresponding Q-value distribution;

The loss function of the Bayesian neural network is as follows:

，

，

In the method, in the process of the invention, Parameters representing bayesian neural networksThe corresponding total loss is calculated and the total loss,Representing a total number of samples in the data sample set; representing a negative log-likelihood loss term, Representing the target Q value of the ith sampleProbability density under a normal distribution predicted by Bayesian neural network, the normal distribution being a mean value corresponding to an ith sampleSum of variancesDescription; indicating the target Q value corresponding to the i-th sample, Representing the mean of the Q-value distribution predicted by the bayesian neural network for the ith sample,Representing the variance of the Q-value distribution of the bayesian neural network for the i-th sample prediction; representing an uncertainty regularization term; Representing regularization coefficients for controlling weights of the regularization terms; Representing the difference of each other Weighting the reciprocal of (2); is the immediate prize of the ith sample, is the corresponding status Down selection actionThe rewards obtained later; is a discount factor for measuring the importance of future rewards; a boolean flag for the ith sample; Indicating the end of the current round of the ith sample, no subsequent state transition actions, when the target Q value is only the instant prize ；Indicating that the current round is not finished, and obtaining rewards when the subsequent state transition actions exist, wherein the target Q value is the current rewardsPlus a maximum expected jackpotDiscount values of (2); Representing execution of an action The new state that is reached after the time,Is shown in a new stateThe potential actions to be taken are as follows,Representing new states of target Q network estimationAnd actionsThe corresponding Q value is used for the control of the temperature,Is a parameter of the target Q network;

the real-time temperature mean is calculated by:

，

In the method, in the process of the invention, Indicating the number of thermocouples in the first thermocouple array,Represent the firstThe real-time temperature values of the individual thermocouples,Representing a real-time temperature average;

The real-time temperature variance is calculated by:

，

In the method, in the process of the invention, The real-time temperature variance is represented and used for measuring the uniformity degree of real-time temperature distribution;

The temperature change rate average is calculated by:

，

In the method, in the process of the invention, The average rate of temperature change is the average rate of temperature change over time; Represent the first The temperature value of each thermocouple at the last sampling time point; representing the time length of a preset time period;

The determining the OTC feature variable according to the real-time temperature average value, the real-time temperature variance and the temperature change rate average value includes:

，

，

，

，

In the method, in the process of the invention, 、AndRespectively representing the normalized real-time temperature mean value, the normalized real-time temperature variance and the normalized temperature change rate mean value; A historical average value representing the average value of the temperature, Represents the historical standard deviation of the temperature mean,Represents the historical mean of the temperature variance,Represents the historical standard deviation of the temperature variance,A historical average representing the average of the rate of change of temperature,The historical standard deviation representing the average value of the temperature change rate is obtained according to the calculation of the historical record of the preset window period; the characteristic variables of the OTC are represented, And the parameter weight coefficients respectively represent corresponding variable parameters.

2. The method of claim 1, wherein prior to determining the real-time OTC temperature based on the temperature signal matrix acquired by the first thermocouple array when the gas turbine is in the no-load free-spinning stage, the method further comprises:

the anti-surge valve closing signal is monitored to identify whether the gas turbine is in an idle free-spinning phase.

3. The method of claim 1 or 2, wherein the first array of thermocouples comprises a first band-responsive thermocouple group and a second band-responsive thermocouple group, each thermocouple in the first band-responsive thermocouple group being uniformly distributed at a first distribution density in a core flow region of the flue gas, and each thermocouple in the second band-responsive thermocouple group being uniformly distributed at a second distribution density in other regions of the first region than the core flow region of the flue gas, wherein the first distribution density is greater than the second distribution density.

4. A method according to claim 3, wherein the parameter weight coefficients of the respective variable parameters are adaptively adjusted by initial value setting and according to a fuzzy rule base and real-time fuel engine performance index feedback; the fuzzy rule base comprises a plurality of fuzzy rules, and each fuzzy rule comprises a fuzzy state of a calibration variable parameter, a calibration fuel engine performance index feedback and a calibration adjustment quantity of a corresponding parameter weight coefficient.

5. A MBR thermocouple based gas turbine OTC temperature optimization system for implementing the MBR thermocouple based gas turbine OTC temperature optimization method of any of claims 1-4; the system comprises:

The temperature acquisition unit is used for determining the real-time OTC temperature based on the temperature signal matrix acquired by the first thermocouple array when the gas turbine is in the idle free rotation stage; the first thermocouple array comprises a plurality of first thermocouples distributed in a first area adjacent to the exhaust port, the first thermocouples adopt wave band response thermocouples, and each first thermocouple respectively has a unique corresponding measurement wavelength;

The variable extraction unit is used for calculating a real-time temperature mean value and a real-time temperature variance according to the real-time OTC temperature, calculating a temperature change rate mean value according to the real-time OTC temperature and a historical OTC temperature group corresponding to a preset time period, and determining an OTC characteristic variable according to the real-time temperature mean value, the real-time temperature variance and the temperature change rate mean value;

An optimization parameter determination unit for inputting the OTC feature variable to a parameter optimization model to determine an optimization control parameter for the gas turbine; the optimal control parameters include at least one of: fuel supply, air supply, inlet guide vane angle, and exhaust temperature setpoint.