A method of monitoring a circulating fluidized bed boiler

Abstract

This invention discloses a monitoring method for a circulating fluidized bed (CFB) boiler, comprising: Step S1, for the CFB boiler to be monitored, pre-acquiring its boiler load at multiple different times during its operational phase, and the corresponding nitrogen oxide (NOx) emission concentration and thermal efficiency for each boiler load; Step S2, pre-establishing a neural network model; training to obtain prediction models for NOx emission concentration and thermal efficiency of the CFB boiler; Step S3, acquiring the boiler load of the CFB boiler at the current time, and then inputting it into the prediction models for NOx emission concentration and thermal efficiency respectively to obtain the NOx emission concentration and thermal efficiency of the CFB boiler at the current time. This invention can efficiently and reliably predict the thermal efficiency and / or NOx emission concentration of a CFB boiler in real time, achieving the purpose of real-time monitoring of the CFB boiler.

Description

Technical Field

[0001] This invention relates to the field of boiler combustion process monitoring technology, and in particular to a monitoring method for a circulating fluidized bed boiler. Background Technology

[0002] Currently, thermal power generation still dominates the electricity supply, with coal being the primary heat source. During the combustion process in power plant boilers, large amounts of polluting gases, such as nitrogen oxides, are produced, causing significant environmental damage. Simultaneously, a large quantity of non-renewable coal is consumed. Therefore, improving the combustion thermal efficiency of circulating fluidized bed boilers in power plants and reducing their pollutant emission concentrations are urgent problems that thermal power generation companies need to address.

[0003] To improve the combustion thermal efficiency of circulating fluidized bed boilers in power plants and reduce their pollutant emission concentrations, it is necessary to predict the thermal efficiency and nitrogen oxide emission concentrations of circulating fluidized bed boilers. Accurate prediction results can help and guide the managers of thermal power plants to deploy boiler equipment operation plans accordingly, thereby ensuring the safe and efficient operation of circulating fluidized bed boilers.

[0004] However, existing methods for predicting the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boilers suffer from poor accuracy and timeliness, failing to meet the needs of users such as thermal power plants.

[0005] Therefore, there is an urgent need to develop a technology that can solve the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this invention is to provide a monitoring method for circulating fluidized bed boilers, addressing the technical deficiencies of existing technologies.

[0007] Therefore, the present invention provides a monitoring method for a circulating fluidized bed boiler, which includes the following steps:

[0008] Step S1: For the circulating fluidized bed boiler to be monitored, obtain in advance the boiler load at multiple different times during the operation phase, as well as the nitrogen oxide emission concentration and thermal efficiency corresponding to each boiler load.

[0009] Step S2: First, establish a neural network model; then, based on the neural network model, train a predictive model to obtain the nitrogen oxide emission concentration of the circulating fluidized bed boiler, and train a predictive model to obtain the thermal efficiency of the circulating fluidized bed boiler.

[0010] Step S3: Obtain the boiler load of the circulating fluidized bed boiler at the current moment in real time, and then input it into the prediction model of nitrogen oxide emission concentration and thermal efficiency of the circulating fluidized bed boiler established in step S2, respectively, to obtain the nitrogen oxide emission concentration and thermal efficiency of the circulating fluidized bed boiler at the current moment.

[0011] As can be seen from the technical solution provided by the present invention above, compared with the prior art, the present invention provides a monitoring method for circulating fluidized bed boilers. Its design is scientific and can efficiently and reliably predict the thermal efficiency and / or nitrogen oxide emission concentration of circulating fluidized bed boilers in real time, so as to achieve the purpose of real-time monitoring of circulating fluidized bed boilers. This is beneficial to help and guide the managers of thermal power plants to deploy the operation plan of boiler equipment accordingly, thereby ensuring the safe and efficient operation of circulating fluidized bed boilers, which has significant practical significance.

[0012] Compared with existing technologies, the novel neural network model proposed in this invention has the following beneficial technical effects:

[0013] 1. The present invention establishes a real-time prediction model for circulating fluidized bed boiler equipment, which can effectively and reliably predict the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boiler equipment in real time.

[0014] 2. Regarding this invention, addressing the complex characteristics of circulating fluidized bed boiler equipment, such as nonlinearity, strong coupling, and large hysteresis, it designs three prediction modules of neural networks with different functions to obtain, as follows: Figure 2 The novel neural network model shown is used to establish a real-time prediction model for the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boilers, and to achieve the purpose of real-time monitoring of the combustion system of circulating fluidized bed boilers.

[0015] 3. Regarding this invention, the novel neural network model proposed in this invention can significantly improve the accuracy of the model for predicting thermal efficiency and nitrogen oxide emission concentration compared to other existing boiler combustion system modeling methods, while ensuring the model's calculation speed. The prediction results obtained through this model are more accurate and timely, thereby achieving the purpose of real-time monitoring of wind power and helping managers of thermal power plants to deploy boiler equipment operation plans in real time.

[0016] 4. In the basic prediction module of the neural network model of this invention, after the input data of the circulating fluidized bed boiler equipment is input from the input layer, it undergoes a nonlinear transformation and linear merging through randomly generated input weights in the hidden layer. Then, the output weights between the hidden layer and the output data are calculated in the output layer. The basic prediction result for the output data can be calculated through the hidden layer and the output weights. This method of nonlinear transformation and linear merging of the input layer is more conducive to extracting effective features from the input data.

[0017] 5. In the reinforcement prediction module of the neural network model of this invention, the steps of the basic prediction module are repeated multiple times to calculate a certain number of basic prediction results with certain differences. These basic prediction results are then linearly merged in the representation layer of the reinforcement prediction module. Finally, the connection weights between the basic prediction results and the output data are calculated in the output layer. Using the basic prediction results and the connection weights, the reinforcement prediction result for the output data can be calculated. This mechanism of stacking and reinforcing basic prediction results using the representation layer is more conducive to improving the stability and generalization ability of the neural network model.

[0018] 6. In the enhanced prediction module of the neural network model of this invention, the residual layer of the enhanced prediction module is obtained by calculating the residual between the enhanced prediction result and the output data of the circulating fluidized bed boiler equipment in the output layer. This residual layer serves as the output layer of both the basic prediction module and the enhanced prediction module. The calculation steps of the enhanced prediction module are repeatedly repeated to obtain a certain number of increasingly smaller enhanced prediction results for the residuals. The enhanced prediction results for the output data and the enhanced prediction results for the residuals are then summed to obtain the enhanced prediction result for the output data. This mechanism of continuously fitting the residuals for gradient enhancement is more conducive to improving the accuracy of the neural network model. Attached Figure Description

[0019] Figure 1 A basic flowchart of a monitoring method for a circulating fluidized bed boiler provided by the present invention;

[0020] Figure 2 A schematic diagram of the topology of a novel neural network model used in a monitoring method for a circulating fluidized bed boiler provided by the present invention.

[0021] Figure 3 A schematic diagram of the fitting curve of the real-time prediction model of the thermal efficiency of a circulating fluidized bed boiler obtained in an embodiment of the monitoring method for circulating fluidized bed boiler provided by the present invention.

[0022] Figure 4 This invention provides a monitoring method for a circulating fluidized bed boiler, and in an embodiment, a schematic diagram of the fitting curve of the real-time prediction model for nitrogen oxides in the circulating fluidized bed boiler is obtained. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0024] See Figures 1 to 4 This invention provides a monitoring method for a circulating fluidized bed boiler, comprising the following steps:

[0025] Step S1: For the circulating fluidized bed boiler to be monitored, the boiler load at multiple different times during the operation phase is obtained in advance, as well as the nitrogen oxide (NOx) emission concentration and thermal efficiency corresponding to each boiler load.

[0026] Step S2: First, establish a neural network model; then, based on the neural network model, train a predictive model to obtain the NOx emission concentration of the circulating fluidized bed boiler, and train a predictive model to obtain the thermal efficiency of the circulating fluidized bed boiler.

[0027] Step S3: Obtain the boiler load of the circulating fluidized bed boiler at the current moment in real time, and then input it into the prediction model of nitrogen oxide (NOx) emission concentration and thermal efficiency of the circulating fluidized bed boiler established in step S2, respectively, to obtain the nitrogen oxide (NOx) emission concentration and thermal efficiency of the circulating fluidized bed boiler at the current moment.

[0028] In step S1, specifically, different times, i.e. different time points, such as the 1st second, 2nd second, 3rd second after the operation begins. ……、 Multiple time points, such as the 100th second.

[0029] In step S1, specifically, the circulating fluidized bed boiler to be monitored is a circulating fluidized bed boiler that has been put into operation, that is, a boiler that has been in operation in the past (previous period).

[0030] It should be noted that for circulating fluidized bed (CFB) boilers, the distributed control system (DCS) integrated with the boiler itself can acquire the boiler load at multiple different times during the boiler's operation phase, as well as the corresponding NOx emission concentration and thermal efficiency for each load. Specifically, the DCS collects data every 2 seconds, resulting in a total of 28,800 data samples. Because the CFB boiler operates smoothly and the sampling interval is small, other characteristics under the same boiler load do not fluctuate significantly. To ensure the model's generalization ability, the 28,800 data samples are resampled every 10 units, increasing the range of variation between data points. Therefore, a total of 2,880 data samples were collected from a 330MW CFB boiler at different times.

[0031] In this invention, it should be noted that there are 26 characteristics that affect the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boilers, including boiler load, coal feed rate, primary air velocity, secondary air velocity, and oxygen concentration in flue gas. Among these, boiler load is the most important and has the greatest impact.

[0032] In step S2, specifically, a predictive model for the nitrogen oxide (NOx) emission concentration of the circulating fluidized bed boiler is trained based on a neural network model, which includes the following operations:

[0033] The boiler load at various times during the operation phase of the circulating fluidized bed boiler is used as the input data for the neural network model, and the nitrogen oxide (NOx) emission concentration corresponding to each boiler load is used as the output data for the neural network model. The neural network model is trained to obtain a prediction model for the NOx emission concentration of the circulating fluidized bed boiler.

[0034] In step S2, specifically, a predictive model for the thermal efficiency of a circulating fluidized bed boiler is trained based on a neural network model, which includes the following operations:

[0035] The boiler load at various different times during the operation phase of the circulating fluidized bed boiler is used as the input data of the neural network model, and the thermal efficiency corresponding to each boiler load is used as the output data of the neural network model. The neural network model is trained to obtain a predictive model of the thermal efficiency of the circulating fluidized bed boiler.

[0036] For the specific implementation of step S2, see [link to relevant documentation]. Figure 2 As shown, the neural network model includes a basic prediction module, a reinforcement prediction module, and an enhancement prediction module that process the input and output data in sequence.

[0037] In this process, the input data (specifically, the boiler load) of the circulating fluidized bed boiler equipment is input from the input layer of the basic prediction module. Then, through the implicit layer of the basic prediction module, the representation layer of the enhanced prediction module, and the residual layer of the improved prediction module, the input data undergoes nonlinear transformation, linear merging, stacking enhancement, and gradient improvement processing operations, respectively. Finally, a mapping relationship is established between the input data and the output data (specifically, the nitrogen oxide emission concentration or thermal efficiency) of the circulating fluidized bed boiler.

[0038] It should be noted that the input layer of the basic prediction module is used to input the input data of the circulating fluidized bed boiler (specifically including the boiler load);

[0039] The hidden layer of the basic prediction module is used to perform nonlinear transformations and linear merging of the input data;

[0040] The representation layer of the enhanced prediction module is used to stack and enhance the input data that has undergone nonlinear transformation and linear merging processing of the hidden layer of the basic prediction module.

[0041] The residual layer of the enhancement prediction module is used to perform gradient boosting on the input data that has been enhanced by stacking the representation layer of the enhancement prediction module.

[0042] In this invention, the basic prediction module specifically includes an input layer, a hidden layer, and an output layer. Input data from the circulating fluidized bed boiler equipment is input through the input layer, then subjected to nonlinear transformation and linear merging through randomly generated input weights in the hidden layer. The output weights between the hidden layer and the output data are then calculated in the output layer. Based on the hidden layer and the output weights, the basic prediction result for the output data can be calculated. The model accuracy of this basic prediction result is typically low, and due to the presence of random weights in the hidden layer, the model stability is usually poor.

[0043] The enhanced prediction module specifically includes an input layer, a hidden layer, and an output layer (which can directly adopt the input, hidden, and output layers of the basic prediction module), as well as a representation layer with special functions. By repeatedly performing the steps of the basic prediction module, a certain number of basic prediction results with some differences can be calculated. These basic prediction results are then linearly merged in the representation layer of the enhanced prediction module. Finally, the connection weights between the basic prediction results and the output data are calculated in the output layer. Using the basic prediction results and the connection weights, the enhanced prediction result for the output data can be calculated. This enhanced prediction result shows improved model accuracy compared to the basic prediction result, and due to the stacked enhancement of multiple basic prediction results implemented in the representation layer, the model stability is relatively strong.

[0044] The enhanced prediction module specifically includes an input layer, a hidden layer, a representation layer, and an output layer (which can directly adopt the input, hidden, representation, and output layers of the enhanced prediction module), as well as a residual layer with special utility. By calculating the residual between the enhanced prediction result and the output data of the circulating fluidized bed boiler equipment in the output layer, the residual layer of the enhanced prediction module is obtained. This residual layer serves as the output layer for both the basic prediction module and the enhanced prediction module. The steps of the enhanced prediction module are repeated continuously to obtain a certain number of increasingly smaller enhanced prediction results for the residuals. The enhanced prediction results for the output data and the enhanced prediction results for the residuals are then summed to obtain the enhanced prediction result for the output data. The model accuracy of this enhanced prediction result is significantly improved compared to the basic prediction result, and its model stability is also typically higher.

[0045] In step S2, specifically, the neural network model is a neural network model with preset calculation rules;

[0046] The preset calculation rules of the neural network model include the following four stages of calculation and processing operations performed sequentially, namely steps S21 to S24. Specifically, the four operations are ordered according to the complexity of their establishment process, as follows:

[0047] Step S21: Perform the computational processing operation from the input layer to the hidden layer of the basic prediction module;

[0048] Step S22: Perform the computational processing operation from the hidden layer to the output layer of the basic prediction module;

[0049] Step S23: Perform the computational processing operation from the representation layer to the output layer of the enhanced prediction module;

[0050] Step S24: Perform the calculation and processing operation from the output layer of the prediction module to the residual layer.

[0051] In this invention, step S21 performs the computational processing operation from the input layer to the hidden layer of the basic prediction module, specifically including the following sub-steps:

[0052] Step S211: The input and output data of the circulating fluidized bed boiler are given in advance. Assume that the input data is X∈R. t×s The output data is Y∈R s×1 ;

[0053] Where, X∈R t×s Y∈R represents the input data of the input layer of the neural network model. s×1 This represents the output data of the output layer of the neural network model;

[0054] In step S211, specifically, the input data is the boiler load of the circulating fluidized bed boiler at multiple different times during the operation phase.

[0055] The output data specifically includes the nitrogen oxide emission concentration or thermal efficiency corresponding to each boiler load.

[0056] Step S212: Randomly generate m×n input weights W ij ∈R s×l and random bias β ij ∈R s×l Then, the result of the nonlinear transformation in the basic prediction module is calculated, and the specific calculation formula is as follows:

[0057]

[0058] in, It is a neural network activation function;

[0059] Where i = 1, 2, ..., m; j = 1, 2, ..., m;

[0060] l is the number of neuron nodes in the nonlinear transformation of the basic prediction module;

[0061] m represents the number of stack reinforcements in the reinforcement prediction module (m is a natural number greater than 0);

[0062] n-1 represents the number of gradient boosting iterations in the prediction module (where n is a natural number greater than 1);

[0063] t is the sample size, and s is the number of features;

[0064] Step S213: Linearly merge the results of the nonlinear transformation in the input layer of the neural network model and the basic prediction module to obtain the hidden layer in the basic prediction module, specifically obtained through the following formula:

[0065]

[0066] In this invention, step S22 performs the computational processing operation from the hidden layer to the output layer of the basic prediction module, specifically including the following sub-steps:

[0067] Step S221: Use the least squares method to calculate the hidden layer in the basic prediction module. Output weights between output layer Y The specific calculation formula is as follows:

[0068]

[0069]

[0070] Step S222: Calculate the basic prediction results of the basic prediction module for the output data of the circulating fluidized bed boiler. The specific calculation formula is as follows:

[0071]

[0072] In this invention, step S23 performs the computational processing operation from the representation layer to the output layer of the enhancement prediction module, specifically including the following sub-steps:

[0073] Step S231: Repeat steps S213 and S22m times (m is a natural number greater than 0) to obtain m basic prediction results. Input the basic prediction results from the representation layer of the enhanced prediction module. The specific representation of the m basic prediction results is as follows:

[0074]

[0075] Step S232: Using the least squares method, calculate the connection weights between the representation layer and the output layer Y of the reinforcement prediction module. The specific calculation formula is as follows:

[0076]

[0077]

[0078] Step S233: Calculate the enhanced prediction result of the enhanced prediction module for the output data of the circulating fluidized bed boiler equipment. The specific calculation formula is as follows:

[0079]

[0080] In this invention, step S24 performs the calculation processing operation from the output layer of the prediction module to the residual layer, specifically including the following sub-steps:

[0081] Step S241: Calculate the enhancement prediction results The residual r between the output layer Y and the output layer Y j The residuals are input from the residual layer of the boosting prediction module, and the specific calculation formula is as follows:

[0082]

[0083] In step S242, the residual layer is used as the output layer of the basic prediction module and the enhanced prediction module. Steps S213, S22 and S23 are repeated to obtain the basic prediction results of the residual data. and enhance prediction results And calculate the residual r j Enhanced prediction results With residual r j The residuals between them (the specific calculation is: residual r)j Enhanced prediction results For residual r j Perform a direct difference operation.

[0084] Step S243: Repeat step S242n-1 times to obtain the enhancement prediction results of n-1 residuals. The specific representation of the enhancement prediction results of n-1 residuals is as follows:

[0085]

[0086] Step S244: Calculate the improvement prediction result of the improvement prediction module for the output data of the circulating fluidized bed boiler equipment. The specific calculation formula is as follows:

[0087]

[0088] In this invention, it should be noted that the parameters in the calculation and processing operations of the neural network model described in steps S21 to S24 are described in Table 1 below:

[0089] Table 1:

[0090] parameter describe t Sample size s Feature number l Number of neurons in the nonlinear transformation of the basic prediction module m Stack reinforcement count in the reinforcement prediction module n-1 Increase the number of gradient boosts in the prediction module

[0091] To better understand the technical solution of the present invention, the following specific embodiments will be used to illustrate the technical solution of the present invention.

[0092] Example.

[0093] First, 26 characteristics affect the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boilers, mainly including boiler load, coal feed rate, primary air velocity, secondary air velocity, and oxygen concentration in flue gas. A total of 2880 data samples were collected from a 330MW circulating fluidized bed boiler at different times, and divided into training and test sets in a 7:3 ratio.

[0094] Secondly, the neural network model proposed in this invention is used to model the thermal efficiency and nitrogen oxide emission concentration of the circulating fluidized bed boiler, respectively. That is, a predictive model for the nitrogen oxide (NOx) emission concentration of the circulating fluidized bed boiler is trained, and a predictive model for the thermal efficiency of the circulating fluidized bed boiler is trained.

[0095] The model parameters are set as follows: the sample size t is set to 2880, the number of features s is set to 26, the number of neurons l in the nonlinear transformation of the basic prediction module is set to 100, the number of stacking boosts m in the boosting prediction module is set to 10, and the number of gradient boosts n-1 in the boosting prediction module is set to 20. In addition, the activation function of the hidden layer in the basic prediction module is the sigmoid function.

[0096] Finally, the fitting curves of the neural network model proposed in this invention for real-time prediction of the thermal efficiency and nitrogen oxide emission concentration of the circulating fluidized bed boiler are plotted as follows: Figure 3 and Figure 4 As shown, this represents a partial fitting result of the neural network model on the thermal efficiency of a circulating fluidized bed boiler. Figure 3 As shown, partial fitting results for nitrogen oxide emission concentrations in circulating fluidized bed boilers are as follows: Figure 4 As shown.

[0097] In this invention, Figure 3 This is a fitting curve for the real-time prediction of the thermal efficiency of a circulating fluidized bed boiler using the neural network model proposed in this invention. The example sample is one-third of the test set data selected randomly, and a portion of the test set data was selected.

[0098] In this invention, Figure 4 The fitted curves for the real-time prediction of nitrogen oxide emission concentrations from circulating fluidized bed boilers using the proposed neural network model are shown. The example samples are one-third of the test set data selected randomly, and a portion of the test set data is also included.

[0099] Depend on Figure 3 and Figure 4 It can be seen that, based on the neural network model proposed in this invention, the output value of the real-time prediction model of the circulating fluidized bed boiler combustion system (i.e., Figure 3 and Figure 4 The predicted value in the model can closely approximate the actual value, thus verifying that the monitoring method for circulating fluidized bed boilers proposed in this invention and the real-time modeling method used therein are effective and reliable technical solutions.

[0100] In summary, compared with existing technologies, the monitoring method for circulating fluidized bed boilers provided by this invention is scientifically designed and can efficiently and reliably predict the thermal efficiency and / or nitrogen oxide emission concentration of circulating fluidized bed boilers in real time, thereby achieving the purpose of real-time monitoring of circulating fluidized bed boilers. This method is beneficial for helping and guiding managers of thermal power plants to deploy boiler equipment operation plans accordingly, thereby ensuring the safe and efficient operation of circulating fluidized bed boilers, and has significant practical implications.

[0101] Compared with existing technologies, the novel neural network model proposed in this invention has the following beneficial technical effects:

[0102] 1. The present invention establishes a real-time prediction model for circulating fluidized bed boiler equipment, which can effectively and reliably predict the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boiler equipment in real time.

[0103] 2. Regarding this invention, addressing the complex characteristics of circulating fluidized bed boiler equipment, such as nonlinearity, strong coupling, and large hysteresis, it designs three prediction modules of neural networks with different functions to obtain, as follows: Figure 2 The novel neural network model shown is used to establish a real-time prediction model for the thermal efficiency and nitrogen oxide emission concentration of circulating fluidized bed boilers, and to achieve the purpose of real-time monitoring of the combustion system of circulating fluidized bed boilers.

[0104] 3. Regarding this invention, the novel neural network model proposed in this invention can significantly improve the accuracy of the model for predicting thermal efficiency and nitrogen oxide emission concentration compared to other existing boiler combustion system modeling methods, while ensuring the model's calculation speed. The prediction results obtained through this model are more accurate and timely, thereby achieving the purpose of real-time monitoring of wind power and helping managers of thermal power plants to deploy boiler equipment operation plans in real time.

[0105] 4. In the basic prediction module of the neural network model of this invention, after the input data of the circulating fluidized bed boiler equipment is input from the input layer, it undergoes a nonlinear transformation and linear merging through randomly generated input weights in the hidden layer. Then, the output weights between the hidden layer and the output data are calculated in the output layer. The basic prediction result for the output data can be calculated through the hidden layer and the output weights. This method of nonlinear transformation and linear merging of the input layer is more conducive to extracting effective features from the input data.

[0106] 5. In the reinforcement prediction module of the neural network model of this invention, by repeating the steps of the basic prediction module, a certain number of basic prediction results with certain differences can be calculated. These basic prediction results are then linearly merged in the representation layer of the reinforcement prediction module. Finally, the connection weights between the basic prediction results and the output data are calculated in the output layer. Using the basic prediction results and the connection weights, the reinforcement prediction result for the output data can be calculated. This mechanism of stacking and reinforcing basic prediction results using the representation layer is more conducive to improving the stability and generalization ability of the neural network model.

[0107] 6. In the enhanced prediction module of the neural network model of this invention, the residual layer of the enhanced prediction module is obtained by calculating the residual between the enhanced prediction result and the output data of the circulating fluidized bed boiler equipment in the output layer. This residual layer serves as the output layer of both the basic prediction module and the enhanced prediction module. The calculation steps of the enhanced prediction module are repeatedly repeated to obtain a certain number of increasingly smaller enhanced prediction results for the residuals. The enhanced prediction results for the output data and the enhanced prediction results for the residuals are then summed to obtain the enhanced prediction result for the output data. This mechanism of continuously fitting the residuals for gradient enhancement is more conducive to improving the accuracy of the neural network model.

[0108] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A monitoring method for a circulating fluidized bed boiler, characterized in that, Includes the following steps: Step S1: For the circulating fluidized bed boiler to be monitored, obtain in advance the boiler load at multiple different times during the operation phase, as well as the nitrogen oxide emission concentration and thermal efficiency corresponding to each boiler load. Step S2: First, establish a neural network model; then, based on the neural network model, train a predictive model to obtain the nitrogen oxide emission concentration of the circulating fluidized bed boiler, and train a predictive model to obtain the thermal efficiency of the circulating fluidized bed boiler. Step S3: Obtain the boiler load of the circulating fluidized bed boiler at the current moment in real time, and then input it into the prediction model of nitrogen oxide emission concentration and thermal efficiency of the circulating fluidized bed boiler established in step S2, respectively, to obtain the nitrogen oxide emission concentration and thermal efficiency of the circulating fluidized bed boiler at the current moment. In step S2, a prediction model for the nitrogen oxide emission concentration of the circulating fluidized bed boiler is trained based on a neural network model, specifically including the following operations: The boiler loads of the circulating fluidized bed boiler at different times during its operation phase are used as input data for the neural network model, and the nitrogen oxide emission concentrations corresponding to each boiler load are used as output data for the neural network model. The neural network model is trained to obtain a prediction model for the nitrogen oxide emission concentrations of the circulating fluidized bed boiler. In step S2, a predictive model for the thermal efficiency of a circulating fluidized bed boiler is trained based on a neural network model, specifically including the following operations: The boiler loads of the circulating fluidized bed boiler at different times during its operation phase are used as input data for the neural network model, and the thermal efficiency corresponding to each boiler load is used as output data for the neural network model. The neural network model is then trained to obtain a predictive model for the thermal efficiency of the circulating fluidized bed boiler. The neural network model includes a basic prediction module, a reinforcement prediction module, and an enhancement prediction module that process the input and output data sequentially. The input data of the circulating fluidized bed boiler equipment is input from the input layer of the basic prediction module. After the input data is processed by the hidden layer of the basic prediction module, the representation layer of the enhanced prediction module, and the residual layer of the improved prediction module, the input data is processed by nonlinear transformation, linear merging, stacking enhancement, and gradient improvement, respectively, to establish a mapping relationship between the input data and the output data of the circulating fluidized bed boiler. A neural network model is a neural network model with pre-defined computational rules. The preset calculation rules of the neural network model include the following four stages of calculation processing operations performed sequentially: Step S21: Perform the computational processing operation from the input layer to the hidden layer of the basic prediction module; Step S22: Perform the computational processing operation from the hidden layer to the output layer of the basic prediction module; Step S23: Perform the computational processing operation from the representation layer to the output layer of the enhanced prediction module; Step S24: Perform the calculation and processing operation from the output layer of the prediction module to the residual layer.

2. The monitoring method for a circulating fluidized bed boiler as described in claim 1, characterized in that, Step S21 performs the computational processing operation from the input layer to the hidden layer of the basic prediction module, specifically including the following sub-steps: Step S211: The input and output data of the circulating fluidized bed boiler are given in advance. Assume that the input data is X∈R. t×s The output data is Y∈R s×1 ; Where, X∈R t×s Y∈R represents the input data of the input layer of the neural network model. s×1 This represents the output data of the output layer of the neural network model; Step S212: Randomly generate m×n input weights W ij ∈R s×l and random bias β ij ∈R s×l Then, the result of the nonlinear transformation in the basic prediction module is calculated, and the specific calculation formula is as follows: in, It is a neural network activation function; Where i = 1, 2, ..., m; j = 1, 2, ..., n; l is the number of neuron nodes in the nonlinear transformation of the basic prediction module; m represents the number of stack reinforcements in the reinforcement prediction module; n-1 represents the number of gradient boosting iterations in the prediction module; t is the sample size, and s is the number of features; Step S213: Linearly merge the results of the nonlinear transformation in the input layer of the neural network model and the basic prediction module to obtain the hidden layer in the basic prediction module, specifically obtained through the following formula:

3. The monitoring method for a circulating fluidized bed boiler as described in claim 2, characterized in that, Step S22 performs the computational processing operation from the hidden layer to the output layer of the basic prediction module, specifically including the following sub-steps: Step S221: Use the least squares method to calculate the hidden layer in the basic prediction module. Output weights between output layer Y The specific calculation formula is as follows: Step S222: Calculate the basic prediction results of the basic prediction module for the output data of the circulating fluidized bed boiler. The specific calculation formula is as follows:

4. The monitoring method for a circulating fluidized bed boiler as described in claim 3, characterized in that, Step S23 performs the computational processing operation from the representation layer to the output layer of the enhanced prediction module, specifically including the following sub-steps: Step S231: Repeat steps S213 and S22m times to obtain m basic prediction results. Input the basic prediction results from the representation layer of the enhanced prediction module. The specific representation of the m basic prediction results is as follows: Step S232: Using the least squares method, calculate the connection weights between the representation layer and the output layer Y of the reinforcement prediction module. The specific calculation formula is as follows: Step S233: Calculate the enhanced prediction result of the enhanced prediction module for the output data of the circulating fluidized bed boiler equipment. The specific calculation formula is as follows:

5. The monitoring method for a circulating fluidized bed boiler as described in any one of claims 2 to 4, characterized in that, Step S24 performs the calculation and processing operation from the output layer of the prediction module to the residual layer, which specifically includes the following sub-steps: Step S241: Calculate the enhancement prediction results The residual r between the output layer Y and the output layer Y j The residuals are input from the residual layer of the boosting prediction module, and the specific calculation formula is as follows: In step S242, the residual layer is used as the output layer of the basic prediction module and the enhanced prediction module. Steps S213, S22 and S23 are repeated to obtain the basic prediction results of the residual data. and enhance prediction results And calculate the residual r j Enhanced prediction results With residual r j The residuals between; Step S243: Repeat step S242n-1 times to obtain the enhancement prediction results of n-1 residuals. The specific representation of the enhancement prediction results of n-1 residuals is as follows: Step S244: Calculate the improvement prediction result of the improvement prediction module for the output data of the circulating fluidized bed boiler equipment. The specific calculation formula is as follows:

Images