An Improved Neural Network Method for Dynamic Modeling of Boiler Combustion System

Abstract

The present invention discloses an improved neural network boiler combustion system dynamic modeling method, comprising the following steps: A. determining the main parameter categories that affect boiler efficiency, extracting data corresponding to the main parameter categories to form an input data set, and determining the corresponding output data set according to the control target; B. constructing a neural network model of the boiler combustion system, and using the input data set and the output data set to train the neural network model; C. extracting PID control parameters in the trained neural network model, and optimizing the PID control parameters according to changes in the state parameters of the boiler combustion system. The present invention can improve the deficiencies of the prior art, realize dynamic modeling of the boiler combustion system, and improve control accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of power plant automation, in particular to an improved neural network boiler combustion system dynamic modeling method. Background Art

[0002] At present, coal-fired power generation is still the main form of electricity supply in my country. In the context of increasingly tight energy supply, it is of great significance to improve the boiler combustion level of the unit. Due to the complex and changeable types of coal in my country, and the many variables such as unit operating conditions and grid load, it is difficult for the unit boiler to operate under the optimal operating conditions during actual operation. In the prior art, there are many feedforward or feedback control methods applied to boiler combustion control, but due to the rapid changes in various parameters during boiler operation, the control accuracy of the pre-set control system will gradually decrease. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide an improved neural network boiler combustion system dynamic modeling method, which can solve the deficiencies of the prior art, realize the dynamic modeling of the boiler combustion system, and improve the control accuracy.

[0004] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows.

[0005] An improved neural network boiler combustion system dynamic modeling method includes the following steps:

[0006] A. Determine the main parameter categories that affect boiler efficiency, extract the data corresponding to the main parameter categories to form an input data set, and determine the corresponding output data set according to the control target;

[0007] B. Construct a neural network model of the boiler combustion system and train the neural network model using input data sets and output data sets;

[0008] C. Extract the PID control parameters in the trained neural network model and optimize the PID control parameters according to the changes in the state parameters of the boiler combustion system.

[0009] Preferably, in step A, the input data set includes total air volume data, fuel volume data, secondary air opening data, and burnout air opening data, and the output data set includes fly ash carbon content data, flue gas temperature data, nitrogen oxide emission data, and boiler efficiency data.

[0010] Preferably, in step B, the neural network model includes an input layer, two hidden layers and an output layer.

[0011] Preferably, in step B, training the neural network model includes the following steps:

[0012] B1. Initialize the neural network model. The weight between the input layer and the first hidden layer is w ij , the weight between the first hidden layer and the second hidden layer is w jk , the weight between the second hidden layer and the output layer is w kl , the threshold of the first hidden layer neuron is a j , the threshold of the second hidden layer neuron is a k , the threshold of the output layer neuron is a l , where i is the i-th neuron in the input layer, j is the j-th neuron in the first hidden layer, k is the k-th neuron in the second hidden layer, and l is the l-th neuron in the output layer. Construct the activation function corresponding to each layer and set the corresponding kernel function in each hidden layer;

[0013] B2. Use the Bayesian formula to calculate the Gaussian distribution of each weight and determine the optimal weight and threshold;

[0014] B3. Use the input data set and output data to train the optimal weights and thresholds, and use the gradient descent method to calculate the minimum solution interval of the loss function;

[0015] B4. In the minimum solution interval, calculate the deviation function between the actual output of the output neuron and the expected output, and obtain the corrected optimal weight and threshold by calculating the extreme point of the deviation function.

[0016] Preferably, in step C, optimizing the PID control parameters comprises the following steps:

[0017] C1. Identify the constraints of the optimization objective function and the state parameters of the boiler combustion system as antigens;

[0018] C2. Generate initial antibodies. Under the control of constraints, randomly generate several groups of initial design vectors as the initial antibody population of the immune system;

[0019] C3. Calculate the fitness and affinity of each antibody in the initial antibody population;

[0020] C4. Select antibodies with an affinity to the antigen greater than a threshold as memory cells and add them to the memory cell pool;

[0021] C5, memory cells produce a new generation of antibody populations through cloning, crossover and mutation;

[0022] C6. Determine whether the termination condition is met. If so, select the antibody with the greatest affinity to the antigen as the optimal solution for the PID control parameters. Otherwise, return to step C3.

[0023] Preferably, in step C4, memory cells added to the memory cell pool are used to perform immune response with antigens, and the memory cells are grouped according to the similarity of the response results.

[0024] Preferably, in step C5, each group of memory cells is first crossovered within the group to ensure that each memory cell has undergone at least one crossover operation, and then the memory cells are reclassified according to the fitness changes of the memory cells before and after the crossover operation, and the memory cells with increased fitness are cloned, and the memory cells with decreased fitness are mutated.

[0025] The beneficial effects brought about by adopting the above technical scheme are: the present invention dynamically simulates the boiler combustion system by establishing a neural network model, and optimizes the PID control parameters by using an immune genetic algorithm, thereby improving the system control accuracy. Among them, by optimizing the training method of weights and thresholds in the neural network model, the amount of calculation of the traditional gradient descent method is effectively reduced. By improving the cloning, crossover and mutation processes of memory cells, the evolution efficiency of the antibody group is accelerated. The dynamic model established by the present invention has high real-time simulation of the boiler combustion system, and can quickly provide feedback according to changes in the operating conditions of the boiler combustion system, so that the staff can make corresponding adjustments to the boiler control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0027] Reference Figure 1 , a specific embodiment of the present invention comprises the following steps:

[0028] A. Determine the main parameter categories that affect boiler efficiency, extract the data corresponding to the main parameter categories to form an input data set, and determine the corresponding output data set according to the control target;

[0029] B. Construct a neural network model of the boiler combustion system and train the neural network model using input data sets and output data sets;

[0030] C. Extract the PID control parameters in the trained neural network model and optimize the PID control parameters according to the changes in the state parameters of the boiler combustion system.

[0031] In step A, the input data set includes total air volume data, fuel volume data, secondary air opening data, and burnout air opening data, and the output data set includes fly ash carbon content data, flue gas temperature data, nitrogen oxide emission data, and boiler efficiency data.

[0032] In step B, the neural network model includes an input layer, two hidden layers and an output layer.

[0033] In step B, training the neural network model includes the following steps:

[0034] B1. Initialize the neural network model. The weight between the input layer and the first hidden layer is w ij , the weight between the first hidden layer and the second hidden layer is w jk , the weight between the second hidden layer and the output layer is w kl , the threshold of the first hidden layer neuron is a j , the threshold of the second hidden layer neuron is a k , the threshold of the output layer neuron is a l , where i is the i-th neuron in the input layer, j is the j-th neuron in the first hidden layer, k is the k-th neuron in the second hidden layer, and l is the l-th neuron in the output layer. Construct the activation function corresponding to each layer and set the corresponding kernel function in each hidden layer;

[0035] B2. Use the Bayesian formula to calculate the Gaussian distribution of each weight and determine the optimal weight and threshold;

[0036] B3. Use the input data set and output data to train the optimal weights and thresholds, and use the gradient descent method to calculate the minimum solution interval of the loss function;

[0037] B4. In the minimum solution interval, calculate the deviation function between the actual output of the output neuron and the expected output, and obtain the corrected optimal weight and threshold by calculating the extreme point of the deviation function.

[0038] In step C, optimizing the PID control parameters includes the following steps:

[0039] C1. Identify the constraints of the optimization objective function and the state parameters of the boiler combustion system as antigens;

[0040] C2. Generate initial antibodies. Under the control of constraints, randomly generate several groups of initial design vectors as the initial antibody population of the immune system;

[0041] C3. Calculate the fitness and affinity of each antibody in the initial antibody population;

[0042] C4. Select antibodies with an affinity to the antigen greater than a threshold as memory cells and add them to the memory cell pool;

[0043] C5, memory cells produce a new generation of antibody populations through cloning, crossover and mutation;

[0044] C6. Determine whether the termination condition is met. If so, select the antibody with the greatest affinity to the antigen as the optimal solution for the PID control parameters. Otherwise, return to step C3.

[0045] In step C4, the memory cells added to the memory cell pool are used to perform an immune response with the antigen, and the memory cells are grouped according to the similarity of the response results.

[0046] In step C5, each group of memory cells is first crossovered within the group to ensure that each memory cell has undergone at least one crossover operation. Then, the memory cells are reclassified according to the fitness changes of the memory cells before and after the crossover operation. The memory cells with increased fitness are cloned, and the memory cells with decreased fitness are mutated.

[0047] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An improved neural network boiler combustion system dynamic modeling method, characterized in that The following steps are involved: A. Determine the main parameter categories that affect boiler efficiency, extract the data corresponding to the main parameter categories to form an input data set, and determine the corresponding output data set according to the control target; the input data set includes total air volume data, fuel volume data, secondary air opening data, and burnout air opening data; the output data set includes fly ash carbon content data, flue gas temperature data, nitrogen oxide emission data, and boiler efficiency data; B. Construct a neural network model of the boiler combustion system and use the input data set and the output data set to train the neural network model; the neural network model includes an input layer, two hidden layers and an output layer; training the neural network model includes the following steps: B1. Initialize the neural network model. The weight between the input layer and the first hidden layer is w ij , the weight between the first hidden layer and the second hidden layer is w jk , the weight between the second hidden layer and the output layer is w kl , the threshold of the first hidden layer neuron is a j , the threshold of the second hidden layer neuron is a k , the threshold of the output layer neuron is a l , where i is the i-th neuron in the input layer, j is the j-th neuron in the first hidden layer, k is the k-th neuron in the second hidden layer, and l is the l-th neuron in the output layer. Construct the activation function corresponding to each layer and set the corresponding kernel function in each hidden layer; B2. Use the Bayesian formula to calculate the Gaussian distribution of each weight and determine the optimal weight and threshold; B3. Use the input data set and output data to train the optimal weights and thresholds, and use the gradient descent method to calculate the minimum solution interval of the loss function; B4. In the minimum solution interval, calculate the deviation function between the actual output of the output neuron and the expected output, and obtain the corrected optimal weight and threshold by calculating the extreme point of the deviation function; C. Extract the PID control parameters in the trained neural network model and optimize the PID control parameters according to the changes in the state parameters of the boiler combustion system; the optimization of the PID control parameters includes the following steps: C1. Identify the constraints of the optimization objective function and the state parameters of the boiler combustion system as antigens; C2. Generate initial antibodies. Under the control of constraints, randomly generate several groups of initial design vectors as the initial antibody population of the immune system; C3. Calculate the fitness and affinity of each antibody in the initial antibody population; C4. Select antibodies with an affinity to the antigen greater than a threshold as memory cells and add them to the memory cell pool; C5, memory cells produce a new generation of antibody populations through cloning, crossover and mutation; C6. Determine whether the termination condition is met. If so, select the antibody with the greatest affinity to the antigen as the optimal solution for the PID control parameters. Otherwise, return to step C3.

2. The improved neural network boiler combustion system dynamic modeling method according to claim 1 is characterized in that: In step C4, the memory cells added to the memory cell pool are used to perform an immune response with the antigen, and the memory cells are grouped according to the similarity of the response results.

3. The improved neural network boiler combustion system dynamic modeling method according to claim 2 is characterized in that: In step C5, each group of memory cells is first crossovered within the group to ensure that each memory cell has undergone at least one crossover operation. Then, the memory cells are reclassified according to the fitness changes of the memory cells before and after the crossover operation. The memory cells with increased fitness are cloned, and the memory cells with decreased fitness are mutated.

Images