CN117892635A - System Simulation and Neural Network Integration Method Based on Modelica - Google Patents

Abstract

The present invention discloses a method for integrating system simulation and neural network based on Modelica. The method uses the Modelica language to construct an accurate system simulation model, obtains key operation data, and uses Python to implement neural network training of key operation data to form a concise and efficient proxy model that can accurately reflect the actual system operation status. The proxy model can be directly used for the function of quickly evaluating the system, reduces the direct dependence on complex Modelica models, and improves computing efficiency. In particular, in scenarios that require frequent operation or rapid evaluation, an accurate and efficient system simulation and analysis method is provided, which has important practical application value for engineering design, system optimization, and decision support.

Description

System Simulation and Neural Network Integration Method Based on Modelica

Technical Field

The invention relates to the field of neural network proxy model integration, and in particular to a method for integrating system simulation and neural network proxy model based on Modelica.

Background technique

Modelica is a modeling and simulation language that is widely used to describe and analyze the physical behavior of dynamic systems. However, as the complexity of system models increases, some models may contain thousands of variables and equations. In this case, traditional simulation methods may face challenges in computing resources and time, especially in scenarios where large-scale parameter optimization or real-time decision-making is required. These challenges may lead to inefficiencies in the simulation process and limit the possibility of in-depth understanding and optimization of system behavior.

As a machine learning technology, the neural network proxy model has the ability to learn models in large-scale data and can be used to accelerate the simulation process of complex systems. Therefore, combining the Modelica simulation model with the neural network proxy model can improve simulation efficiency while maintaining accuracy, especially in scenarios where frequent model prediction and optimization are required. By combining the physical modeling capabilities of Modelica and the data learning capabilities of neural networks, we can better cope with the challenges of complex system modeling and optimization and improve the efficiency of engineering design and decision-making.

Summary of the invention

The present invention proposes a method for integrating system simulation and neural network proxy model based on Modelica. The method constructs a neural network proxy model by using the operation result data of Modelica system simulation, which not only improves the simulation efficiency, but also provides a powerful tool for model simplification and optimization, bringing new possibilities for in-depth understanding and efficient design of complex systems.

A method for integrating system simulation and neural network agent model based on Modelica comprises the following steps:

Step S1, build a Modelica simulation model of the steam cycle system based on the ThermalPower library. The model includes a condensate pump, a steam-water separator, a boiler, a steam-water pump, a superheater, a steam valve, a steam turbine, a condenser and a generator. First, the boiler feed water is heated and pumped into the steam-water separator by the steam-water pump. The water vapor in the steam-water separator is further heated by the superheater, flows through the steam valve for throttling and pressure control, and then enters the steam turbine to expand and generate electricity for the generator. Finally, the steam enters the condenser and condenses into water, which is then transported to the steam-water separator by the condensate pump to complete the cycle.

Step S2, perform system simulation calculations, obtain characteristic data as data set 1, the boiler heating capacity, superheater superheat and condenser pump feed water volume in the simulation results are used as input data of data set 1, and the generator power generation, condenser heat exchange and turbine shaft power are used as observation data of data set 1.

Step S3, normalize the data set 1, and then set the parameters required by the neural network, including the number of samples, input dimension, output dimension, learning rate and number of nodes;

Step S4, training the neural network model, when the error between the model output result and the observed data of data set 1 is less than the error setting value, the prediction agent model is obtained.

Step S5: Use the Modelica system model in step S1 to obtain steam cycle system data under another working condition as data set 2 for verification, input the input data of data set 2 into the prediction proxy model, and compare the output data of the prediction proxy model with the observed data of data set 2. If the error is less than 3%, the proxy model is successfully established; otherwise, return to step S4 and retrain.

In summary, the present invention first uses the Modelica language to build a physical system model based on the process flow and thermodynamic principles. Taking the steam cycle system as an example, a basic thermodynamic cycle system including components such as a boiler, a superheater, a steam turbine, a generator, a condenser, and a pump is constructed, and simulation is performed to obtain detailed operation result data. On this basis, by selecting the characteristic data in the system model as input data and observation data, the input data becomes the input layer of the neural network, and the observation data is the comparison value of the output result of the neural network. The model is trained until the error reaches the predetermined requirement. Thereafter, another operating condition of the system simulation is selected for verification, and finally a simplified and efficient neural network proxy model is generated. Specifically, the present invention generates a refined proxy model by selecting the system simulation result data for neural network training. In the process of training the model, the system is continuously optimized to reduce the error value to achieve the required accuracy level. Through the continuous iterative optimization process, the present invention realizes the efficient simplification of complex system simulation models, and provides an innovative method for faster and more accurate system simulation calculations.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for integrating system simulation and neural network proxy model based on Modelica according to the present invention.

FIG2 is a flow chart of a training model of a method for integrating a system simulation and a neural network proxy model based on Modelica provided in an embodiment of the present invention.

FIG3 is an input data diagram of a data set 1 of a method for integrating system simulation and a neural network proxy model based on Modelica provided in an embodiment of the present invention.

FIG. 4 is a diagram showing output data of a prediction proxy model and observation data of a data set 1 in a method for integrating a system simulation and a neural network proxy model based on Modelica provided by an embodiment of the present invention.

5 is an error diagram of the output data of the prediction proxy model and the observation data of data set 1 of a method for integrating system simulation and neural network proxy model based on Modelica provided by an embodiment of the present invention.

FIG6 is an input data diagram of data set 2 under verification conditions of a method for integrating system simulation and neural network proxy model based on Modelica provided in an embodiment of the present invention.

7 is a diagram of the observed data of data set 2 and the output data of the proxy model under the verification condition of a method for integrating system simulation and neural network proxy model based on Modelica provided by an embodiment of the present invention.

Detailed ways

In order to more clearly understand the subject, technical solutions and advantages of the present invention, the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments. It should be noted that the specific embodiments described below are only for explaining the present invention, but are not limited to the present invention. In addition, the embodiments of the present invention and the technical features in the embodiments can be combined with each other as long as they do not conflict with each other.

The present invention provides a coal-fired boiler efficient operation method based on Modelica modeling and genetic algorithm, which mainly includes the following process:

Step 1, as shown in Figure 1, a Modelica simulation model of a steam cycle system is established based on the ThermalPower library. The model includes a condenser pump, a steam-water separator, a boiler, a steam-water pump, a superheater, a steam valve, a steam turbine, a condenser, and a generator. The connection method is: the boiler feed water is heated and pumped into the steam-water separator through a steam-water pump. The water vapor in the steam-water separator is further heated by the superheater, flows through the steam valve for throttling and pressure control, and then enters the steam turbine to expand and generate electricity for the generator. Finally, the steam enters the condenser and condenses into water, which is then transported to the steam-water separator by the condenser pump to complete the cycle.

Step 2: The simulation duration is set to 1 week (168 hours) with a step length of 1 hour. The system model is simulated and solved using the Dassl solver. Data set 1 selects system simulation result data every hour. The boiler heating supply, superheater superheat and condensing pump feed water volume are used as input data in data set 1. The generator power generation, condenser heat exchange and turbine shaft power are used as observation data in data set 1.

Step 3: Normalize the input data and the observed data, and map the input data and the observed data to the range of (-1, 1).

First, find the minimum value of the input data (boiler heat supply, superheater superheat and water pump feed) and the observed data (generator power generation, condenser heat exchange and turbine shaft power) in data set 1. and maximum value/> , where/> With/> is the minimum and maximum value of the input data, i is the input data dimension, /> With/> is the minimum and maximum value of the observed data, j is the dimension of the observed data; then it is normalized by the following formula:

in, With/> is the data set after the input data and observation data are normalized,/> and are the input data and the observed data.

Then configure the parameters required by the neural network. Each input and output data contains 168 values, so the number of samples is 168. The input data includes boiler heat supply, superheater superheat, and water pump feed water, so the input dimension is 3. The output data includes generator power generation, condenser heat exchange, and turbine shaft power, so the output dimension is 3.

If the learning rate is too high, the model may not converge during the training process, and the update step of the parameters is too large, causing the weights and biases to oscillate back and forth in the parameter space, and ultimately failing to find the optimal solution; if the learning rate is too low, the step size of the model parameter update is too small, and the training process may take a long time to converge to the minimum value and may fall into a local optimal situation. In this example, the learning rate is selected as 0.001. Regarding the number of nodes, if the number of hidden layer nodes is set too high, the model may overfit the training data, that is, it learns the noise and details in the training data, but the generalization performance of new data is poor and the computational complexity increases. If the number of hidden layer nodes is set too low, the model may not be able to capture the complex patterns of the training data, resulting in underfitting and poor accuracy. Therefore, the number of nodes is selected as 10 in this example.

Step 4: As shown in FIG. 2, start training the neural network model. The specific process is as follows:

Step (1), first determine the relationship between the hidden layer and the input layer:

in, is the output of the hidden layer, /> is the standardized input data, i.e., the one in step 3/> , With/> is the weight coefficient, and the initial value is set randomly.

Step (2), calculate the output of the output layer:

in, is the output of the output layer, /> With/> is the weight coefficient, and the initial value is set randomly.

Step (3), calculate the error between the output layer and the observed data:

in, is the overall error, /> To calculate the sum, if /> If the error is smaller than the expected range, stop training; otherwise, execute step (4).

Step (4), update weight coefficient 、/> 、/> 、/> :

The loss function between the input layer and the hidden layer is:

The loss function between the hidden layer and the output layer is:

Update the weight coefficients between the input layer and the hidden layer:

Where rate is the learning rate in step (3);

Update the weight coefficients of the hidden layer and the output layer:

After obtaining the new weight coefficient, execute step (1) and repeat the calculation until the error between the output of the model and the observed data of data set 1 is less than the error setting value of 0.5, and the training ends. Figure 5 shows the error between the output data of the predicted proxy model and the observed data of data set 1 as the number of training times increases. When the training model reaches 17857 times, the error value is 0.49998, which meets the error setting value requirement, and the training stops.

Step 5: Use the Modelica system model in step 1 to obtain steam cycle system data under another working condition as data set 2 for verification. The input data of data set 2 is shown in Figure 6. The input data is input into the prediction agent model. The output data of the prediction agent model is compared with the observation data of data set 2, as shown in Figure 7. The errors of the three observation data are calculated by the following formula:

in, is the sample size, /> Output value of the proxy model under the verification condition, /> The output value of the simulation result of the Modelica system under the verification condition. Observation data: The error values of the power generation of the generator, the heat exchange of the condenser and the shaft power of the turbine are 1.20%, 1.18% and 0.95% respectively. The errors are all less than 3%, and the proxy model is successfully established.

The application of this method in the steam cycle system can simplify the model and improve the calculation efficiency by analyzing the characteristic data and determining the input and output function relationship. In addition, the establishment of the proxy model in this method can also be extended to other system simulation processes or practical application projects, and the goal of quickly establishing a simple model can be achieved by clarifying the input and output data.

The present invention establishes a steam cycle system model by using the ThermalPower library in Modelica, takes the boiler heat supply, superheat of the superheater and the water supply of the water pump as input data, the power generation of the generator, the heat exchange of the condenser and the shaft power of the turbine as observation data, and uses neural network training to successfully obtain a simplified and efficient proxy model. First, the characteristic data in the system simulation results are selected as the input and observation data for neural network training. After normalization, the model training is started until the preset error standard is met. Then the Modelica simulation result data of another working condition is used for verification to ensure that the error of the observation data is less than 3%, and finally a high-precision proxy model is generated. This technology achieves an improvement in computing efficiency, and can provide fast feedback in real-time decision-making, reducing dependence on traditional complex system simulation software, accelerating the system design and optimization process, and quickly predicting the performance of the system under different conditions, and has important engineering application value.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (5)

1. A system simulation and neural network integration method based on Modelica, characterized by comprising the following steps: Step S1, based on the ThermalPower library, a Modelica simulation model of a steam cycle system is established. The model includes a condensate pump, a steam-water separator, a boiler, a steam-water pump, a superheater, a steam valve, a steam turbine, a condenser and a generator. First, the boiler feed water is heated and pumped into the steam-water separator through a steam-water pump. The water vapor in the steam-water separator is further heated by the superheater, flows through the steam valve for throttling and pressure control, and then enters the steam turbine for expansion to generate electricity for the generator. Finally, the steam enters the condenser and condenses into water, which is then transported to the steam-water separator by the condensate pump to complete the cycle. Step S2, perform system simulation calculation, obtain characteristic data as data set 1, the boiler heat supply, superheat of superheater and water supply of condenser pump in the simulation result are used as input data of data set 1, the power generation of generator, heat exchange capacity of condenser and shaft power of steam turbine are used as observation data of data set 1; Step S3, normalize the data set 1, and then set the parameters required by the neural network, including the number of samples, input dimension, output dimension, learning rate and number of nodes; Step S4, training the neural network model, when the error between the model output result and the observed data of data set 1 is less than the error setting value, obtaining the prediction agent model; Step S5, using the Modelica system model in step S1 to obtain the steam cycle system data under another working condition as data set 2, and verify it, input the input data of data set 2 into the prediction proxy model, and compare the output data of the prediction proxy model with the observation data of data set 2. If the error is less than 3%, the proxy model is successfully established; otherwise, return to step S4 and retrain. 2. The method for system simulation and neural network integration based on Modelica according to claim 1 is characterized in that the data set in step S3 is normalized, including input data and observation data, and the specific implementation process of the normalization is as follows: (1) First, find the minimum value of the input data and the observed data in data set 1 and maximum value/> , where/> With/> is the minimum and maximum value of the input data, i is the input data dimension, /> With/> is the minimum and maximum value of the observed data, j is the dimension of the observed data; (2) Map the input data and observation data to the range of (-1, 1): , , in, With/> is the data set after the input data and observation data are normalized,/> With/> are the input data and the observed data. 3. The method for system simulation and neural network integration based on Modelica according to claim 1 is characterized in that the parameters required for the neural network in step S3, including the number of samples, input dimension, output dimension, learning rate and number of nodes, are flexibly customized according to specific application requirements to achieve model performance. 4. The method for system simulation and neural network integration based on Modelica according to claim 1, characterized in that the specific process of training the neural network model in step S4 is as follows: Step (1), first determine the relationship between the hidden layer and the input layer: , in, is the output of the hidden layer, /> is the standardized input data, /> With/> is the weight coefficient; Step (2), calculate the output of the output layer: , in, is the output of the output layer, /> With/> is the weight coefficient; Step (3), calculate the error between the output layer and the observed data: , , in, is the overall error, /> To calculate the sum, if /> If the error is less than the expected error range, stop training, otherwise execute step (4); Step (4), update weight coefficient 、/> 、/> 、/> : The loss function between the input layer and the hidden layer is: , The loss function between the hidden layer and the output layer is: , Update the weight coefficients between the input layer and the hidden layer: , , Where rate is the learning rate in step (3); Update the weight coefficients of the hidden layer and the output layer: , , After obtaining the new weight coefficient, execute step (1) and repeat the cycle calculation again. 5. The method for system simulation and neural network integration based on Modelica according to claim 1, wherein the calibration error value in step S5 is calculated by the following equation: , in, is the sample size, /> Output value of the proxy model under the verification condition, /> It is the output value of the simulation result of Modelica system under the verification condition.