A boiler intelligent combustion optimization method based on module-level mechanism and mathematical hybrid model

Abstract

The present invention relates to a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model, which belongs to the field of boiler combustion technology and solves the technical problems that the existing boiler combustion system model is inaccurate and the optimization effect of the traditional control optimization algorithm is not obvious. The solution is: a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model, comprising the following steps: 1) splitting the boiler system into several modules; 2) using physical information fusion network technology to model the modules that cannot be modeled using a clear mechanism; building a mechanism model skeleton for the modules with clear mechanisms but unclear key parameters; 3) calibrating the real-time data of the boiler; 4) obtaining other boiler combustion data that cannot be directly obtained; 5) designing optimization objectives and optimization decision variables; 6) boiler combustion optimization includes intelligent recommendation, intelligent optimization and safety verification. Compared with the prior art, the present invention has the advantages of improved accuracy of the boiler combustion system model and more obvious intelligent optimization effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of boiler combustion, and in particular relates to a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model. Background Art

[0002] As the main body of my country's power generation capacity, thermal power units are still the main source of regulation capacity in most regions of my country. The promotion of dual carbon goals and new energy strategies requires deep peak regulation units to frequently switch between conventional conditions and low-load conditions, which results in multi-condition and multi-modal characteristics of unit operation. The large lag characteristics of the boiler cause the change process of the boiler to always lag behind the change process of the turbine during deep peak regulation. Conventional control schemes cannot cope with the time lag characteristics of the system, resulting in a low load rise and fall rate of the unit, which cannot meet the response requirements of rapid deep peak regulation, and is prone to over-temperature, over-pressure, and overshoot problems during operation. Optimizing the combustion operation mode of coal-fired boilers in thermal power plants is a key link in improving the unit's combustion thermal efficiency, improving the ability to quickly change loads, and ensuring the safe and stable operation of the unit. Therefore, it is necessary to make advance predictions of process parameters such as boiler pressure, main steam temperature, NOx emissions, and SO2 emissions, and to achieve advance adjustment of the boiler based on the predicted data, so that the boiler matches the rapid change characteristics of the turbine.

[0003] However, the boiler combustion system has the characteristics of large inertia, large time delay, nonlinearity, and multiple disturbances. It is difficult to establish an accurate boiler combustion system model based on traditional mechanism modeling and small data verification, and traditional control optimization algorithms are also difficult to achieve obvious results. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art and solve technical problems such as the inaccurate existing boiler combustion system model and the insignificant optimization effect of the traditional control optimization algorithm, the present invention provides a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model.

[0005] The present invention is achieved through the following technical solutions.

[0006] The present invention provides a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model, which is divided into a boiler key indicator prediction model design and a boiler intelligent combustion optimization model design, and includes the following steps, wherein steps 1) to 2) are boiler key indicator prediction model design, and steps 3) to 6) are boiler intelligent combustion optimization model design:

[0007] 1) Based on the boiler system process and mechanism, the boiler system is divided into air preheating module, coal grinding module, coal quality analysis module, slag analysis module, boiler combustion module, economizer module, water wall module, superheater module, high pressure cylinder module, high pressure heating module, reheater module and medium pressure cylinder module;

[0008] 2) Modeling the modules that cannot be modeled using a clear mechanism in step 1) using physical information fusion network technology; First, build a mechanism model skeleton for the modules with clear mechanisms but unclear key parameters in step 1), and then use neural network technology to fit and correct the model coefficients in real time;

[0009] Based on the boiler module model and the boiler operation history data, a prediction model for key parameters of the boiler is established through prediction model input feature screening, prediction model performance optimization and prediction model optimization; the prediction model input feature screening uses grey correlation and Spearman correlation coefficient algorithms to screen out several parameters with the highest correlation with the quantity to be predicted from a large number of parameters, and further screens parameters with physical significance based on manual experience, and uses the selected parameters as input feature quantities of the prediction model; further analyzes the correlation between the historical time series data of the quantity to be predicted and the current moment data, and selects the top N time series values ​​with the correlation value as the input of the time series prediction model; the prediction model performance optimization is optimized through model feature screening and variance analysis; the prediction model optimization is to select a model with the highest prediction accuracy from several prediction models of a certain key parameter as the prediction model to perform real-time prediction of the key parameter; the key parameters predicted by the prediction model are used as the input of the boiler intelligent combustion optimization model;

[0010] 3) Use upper and lower threshold measures, mutation detection measures, multi-point anomaly detection measures, regression detection measures, energy conservation and physical property equation calibration measures to calibrate the real-time boiler data collected by various sensors as input data for the boiler intelligent combustion optimization model;

[0011] 4) At the same time, the boiler combustion data that cannot be directly obtained is obtained through manual sampling and offline test data entry;

[0012] 5) Design the optimization objectives and optimization decision variables of the boiler combustion optimization model;

[0013] 6) Optimizing boiler combustion according to the optimization target and optimization decision variables of boiler combustion, wherein the boiler combustion optimization includes intelligent recommendation, intelligent optimization and safety verification;

[0014] The intelligent recommendation is to establish a data optimization and clustering model based on boiler efficiency monitoring indicators, boiler combustion efficiency and historical data, and use physical information fusion network technology (PINN) to build the framework of the intelligent recommendation model;

[0015] The intelligent optimization is to establish an algorithm fitness function according to the optimization target, and then use the intelligent recommendation result as the initial value calculation of the genetic algorithm (GA algorithm), non-dominated sorting genetic algorithm, adaptive particle swarm algorithm and several swarm optimization algorithms and select the optimal value to obtain the optimal solution of intelligent optimization;

[0016] The safety check is performed by checking whether the variables are within the specified range and whether the constraints are met. The optimization results are safety checked to verify whether the safety boundaries and constraints are met.

[0017] Furthermore, in the physical information fusion network technology in step 2), an objective function L = λ*Loss(y,y pred )+(1-λ)*║F(x)║ 2 Complete the modeling, where L is the overall objective function, Loss is the loss function of the neural network, y is the observed value, y pred is the predicted value of the neural network, F(x) is the residual of the physical equation, and λ is the parameter that weighs the loss function and the physical equation, which is set to 0.5.

[0018] Further, the Spearman correlation coefficient method in step 2) is based on Calculate, where R i , S i is the i-th variable in the two sets of data after correlation sorting, is the average of the two sets of data, and N is the amount of data.

[0019] Furthermore, the upper and lower threshold measures in step 3) are used to determine whether single-point data is abnormal; the mutation detection measure is based on the ARIMA time series prediction algorithm combined with the 3-sigma principle to determine data reliability; the multi-point anomaly detection measure is based on the PCA and AAKR signal reconstruction algorithm to detect the high-dimensional spatial distribution information of multi-sensor variables and local data anomaly information; the regression detection measure is based on LSTM regression detection for feature collection and preprocessing; the energy conservation and physical property equation calibration measure is used to calibrate information.

[0020] Furthermore, the specific steps of the mutation detection measure are: first, preprocess the historical data of boiler combustion, and then establish a model based on the ARIMA time series prediction algorithm for the preprocessed data, and then use the established model to predict the latest boiler combustion data and use the 3-sigma principle to judge the data reliability. If the predicted value is not within the reliable data range, the data is judged to be mutation data; if the predicted value is within the reliable data range, the data is judged to be reliable data.

[0021] Furthermore, the framework of building the intelligent recommendation model in step 6) includes the following steps:

[0022] Step 1: Preprocess boiler combustion operation data through outlier detection, missing value filling and feature selection;

[0023] Step 2: Determine the boiler efficiency monitoring index, and perform weighted processing on the boiler efficiency monitoring index to obtain a comprehensive efficiency evaluation index;

[0024] Step 3: Based on the boiler efficiency monitoring indicators and boiler combustion history data, a data optimization model is established by the least squares method and the boiler efficiency is evaluated by using a clustering algorithm;

[0025] Step 4: Based on the boiler combustion history data, select the operation strategy with high frequency and confidence by using the association rule mining algorithm and setting the minimum support and minimum confidence parameters;

[0026] Step 5: Obtain the PINN neural network model based on the boiler combustion history data and the PINN neural network structure;

[0027] Step 6: Input the real-time boiler combustion data into the PINN neural network model to predict boiler efficiency.

[0028] Further, the GA algorithm in step 6) includes the following steps:

[0029] Step I: Initialize the population: input the intelligent recommendation results into the GA algorithm population as the initial value;

[0030] Step II: Selection: Select individuals with higher fitness as parents according to the algorithm fitness function;

[0031] Step III: Crossover: Cross the parent generation to generate new offspring;

[0032] Step IV: Mutation: Mutate the offspring;

[0033] Step V: Update the population: Use the parent and child generations to update the population;

[0034] Step VI: Iterate the above steps until the termination condition is met;

[0035] Step VII: Decoding: Obtain the optimal variable value based on the optimal solution.

[0036] Furthermore, in step I, the population size of the GA algorithm is 100, the crossover probability in step III is 0.7, the mutation probability in step IV is 0.001, and the iteration in step VI is 1000 times.

[0037] The beneficial effects achieved by the present invention are as follows: the present invention proposes a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model. Based on the boiler system process and mechanism, the boiler system is divided into several modules. For modules that cannot be modeled using a clear mechanism, physical information fusion network technology is used to model the modules. For modules with clear mechanisms but unclear key parameters, a mechanism is used to build a model skeleton, and a neural network is used to fit and real-time correct the model coefficients. Module modeling further improves the accuracy of the model; an intelligent optimization method of intelligent recommendation, intelligent optimization and safety verification is used to improve boiler efficiency, and the intelligent optimization effect is more obvious.

[0038] Compared with the prior art, the present invention has the advantages of improved accuracy of boiler combustion system model and more obvious intelligent optimization effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a structural block diagram of the present invention;

[0040] Figure 2 is a boiler modular system of the present invention;

[0041] Figure 3 It is the GA algorithm flow chart of the present invention. DETAILED DESCRIPTION

[0042] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0043] like Figures 1 to 3 As shown, a boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model is divided into a boiler key indicator prediction model design and a boiler intelligent combustion optimization model design, including the following steps, wherein steps 1) to 2) are boiler key indicator prediction model design, and steps 3) to 6) are boiler intelligent combustion optimization model design:

[0044] 1) Based on the boiler system process and mechanism, the boiler system is divided into air preheating module, coal grinding module, coal quality analysis module, slag analysis module, boiler combustion module, economizer module, water-cooled wall module, superheater module, high-pressure cylinder module, high-pressure heating module, reheater module and medium-pressure cylinder module. The module modeling further improves the accuracy of the prediction model.

[0045] 2) Use physical information fusion network technology to model the modules that cannot be modeled using clear mechanisms in step 1), such as the boiler combustion module. The internal structure of the boiler, the boiler combustion mechanism, and the radiation heat transfer are difficult to quantify, which makes it impossible to accurately and efficiently establish the boiler model using conventional mechanism modeling methods. In this way, the physical information fusion network technology is used to model the module in a targeted manner to improve the accuracy and generalization performance of the model. Physical information fusion network technology mainly allows physical equations to participate in the modeling process. In this way, the neural network optimizes not only the network's own loss function during training iterations, but also the difference in each iteration of the physical equation, so that the final training result satisfies the laws of physics. The neural network contains two fully connected hidden layers. The number of neurons in each hidden layer is selected to be 64 neurons according to the complexity of the problem, and the activation function is the Sigmoid function. By establishing the objective function L=λ*Loss(y,y pred )+(1-λ)*║F(x)║ 2 Complete the modeling, where L is the overall objective function, Loss is the loss function of the neural network, y is the observed value, y pred is the predicted value of the neural network, F(x) is the residual of the physical equation, and λ is the parameter for weighing the loss function and the physical equation, which is set to 0.5. First, build the mechanism model skeleton for the module with clear mechanism and unclear key parameters in step 1), and then use neural network technology to fit and correct the model coefficients in real time. Taking the superheater module as an example, the implementation process is as follows: first build the mechanism model skeleton, assuming that the heat transfer equation of the superheater is Q=U*A*(ΔT m ), where Q is the heat transfer rate of the superheater, U is the superheat coefficient, A is the heat transfer area, ΔT m is the average temperature difference; secondly, the neural network is used to fit and correct the superheat coefficient in real time, such as U corrected =U initial +f(x), where U corrected is the corrected superheat coefficient, U initial is the initial superheat coefficient, f(x) is the output of the superheat coefficient neural network; finally, the neural network optimization is used to adjust the weight and bias of the neural network by minimizing the loss function Loss combined with the gradient descent optimization algorithm to improve its prediction ability. By inputting the new process parameter x, the modeled neural network can be used to predict the corrected superheat coefficient U corrected , and then obtain a more accurate heat transfer rate Q.

[0046] Based on the boiler module model and the boiler operation history data, a prediction model for key parameters of the boiler is established through prediction model input feature screening, prediction model performance optimization and prediction model optimization; the prediction model input feature screening uses grey correlation and Spearman correlation coefficient algorithms to screen out several parameters with the highest correlation with the quantity to be predicted from a large number of parameters, and further screens parameters with physical significance based on manual experience, and uses the selected parameters as the input feature quantity of the prediction model, wherein the Spearman correlation coefficient method is based on Calculate, where R i , S i is the i-th variable in the two sets of data after correlation sorting, is the average of two sets of data, and N is the amount of data. Further analyze the correlation between the historical time series data of the quantity to be predicted and the current moment data, and select the top N time series values ​​of the correlation value as the input of the time series prediction model. The performance optimization of the prediction model is performed through the feature screening and variance analysis of the model. The prediction model optimization is to select a model with the highest prediction accuracy from several prediction models of a certain key parameter as the prediction model to perform real-time prediction of the key parameter, and the key parameters predicted by the prediction model are used as the input of the boiler intelligent combustion optimization model. Taking NO X For example, the emission prediction is used as an example. The process of establishing the prediction model is as follows: first, the prediction model input feature screening is performed, and then the NO X The relevant parameters generated mainly include unit load, total coal quantity, moisture, ash, volatile matter, fixed carbon, sulfur content, etc. of the coal quality entering the furnace, exhaust oxygen content, primary air flow, secondary air damper opening, peripheral air damper opening, burnout air damper opening and other variables. Then, the Spearman correlation coefficient method is used to analyze the correlation between each variable and the NOx concentration at the inlet of the SCR reactor, and the top N parameters of the correlation are input. Through correlation analysis, unit load, total air volume, total coal quantity, coal feed, exhaust oxygen content, secondary air damper opening, primary air flow and other variables are finally selected as the input features of the prediction model. In addition, NO X The generation of is also related to its state at several previous moments. Secondly, the prediction model is designed. The BILSTM model adopts a bidirectional time transfer time step to obtain a more global time series information by running LSTM in the forward and reverse time directions respectively. The constructed BILSTM model is a 4-layer neural network, including input layer, BILSTM layer, fully connected layer and output layer. The Adam algorithm is used to update the model weights when the model is constructed. In order to prevent gradient explosion, the gradient threshold is set to 1. The model integrates all the information of historical and future data at the same time, which can effectively improve the prediction accuracy and generalization ability of the model.

[0047] 3) Use upper and lower threshold measures, mutation detection measures, multi-point anomaly detection measures, regression detection measures, energy conservation and physical property equation calibration measures to calibrate the real-time boiler data collected by various sensors as input data for the boiler intelligent combustion optimization model. The upper and lower threshold measures are used to determine whether single-point data is abnormal. The mutation detection measures are based on the ARIMA time series prediction algorithm combined with the 3-sigma principle to determine data reliability. The specific steps of the mutation detection measures are: first, pre-process the boiler combustion history data, and then establish a model based on the ARIMA time series prediction algorithm for the pre-processed data, and then use the established model to predict the latest boiler combustion data and use the 3-sigma principle to determine data reliability. If the predicted value is not within the reliable data range, the data is judged to be mutation data; if the predicted value is within the reliable data range, the data is judged to be reliable data. The multi-point anomaly detection measure is based on PCA and AAKR signal reconstruction algorithms to detect high-dimensional spatial distribution information and local data anomaly information of multi-sensor variables. The implementation method is as follows: first, the data of multi-sensor variables is preprocessed by removing noise, filling missing values, and standardizing data; secondly, the preprocessed data is subjected to PCA principal component analysis, and the principal components of the data are obtained by calculating the covariance matrix and eigenvalue decomposition, and then the number of principal components retained is selected to retain sufficient data variance information; then the AAKR signal is used to reconstruct the original data in the principal component space after dimensionality reduction. The AAKR signal reconstruction algorithm considers the correlation of data in process, mechanism, and logic, and uses adaptive K nearest neighbor regression to reconstruct data, and further analyzes the distribution characteristics of the reconstructed data in high-dimensional space, such as the density distribution and clustering of data; finally, combined with the correlation of data in process, mechanism, and logic, the local data is judged for abnormal information. The regression detection measure is based on LSTM regression detection for feature quantity collection and preprocessing; the energy conservation and physical property equation calibration measure is used for calibration information. These measures have improved the accuracy of the data and are further conducive to the optimization of the intelligent optimization model.

[0048] 4) At the same time, the boiler combustion data that cannot be directly obtained is obtained through manual sampling and offline test data entry.

[0049] 5) Design the optimization objectives and optimization decision variables of the boiler combustion optimization model. The boiler combustion operation goal is to improve boiler efficiency, reduce the power consumption of equipment such as induced draft fans and coal mills, and increase the net output of the unit. Operation control optimization decision variables include direct or indirect control variables such as coal mills, primary air, secondary air, and flue gas dampers. By optimizing the objectives and optimization decision variables, potential operation risks such as equipment fatigue and damage can be further avoided.

[0050] 6) Optimizing boiler combustion according to the optimization target and optimization decision variables of boiler combustion, wherein the boiler combustion optimization includes intelligent recommendation, intelligent optimization and safety verification;

[0051] The intelligent recommendation is to establish a data optimization and clustering model based on boiler efficiency monitoring indicators, boiler combustion efficiency and historical data, and use the PINN neural network to build the skeleton of the intelligent recommendation model; wherein building the skeleton of the intelligent recommendation model includes the following steps:

[0052] Step 1: Preprocess boiler combustion operation data through outlier detection, missing value filling and feature selection;

[0053] Step 2: Determine the boiler efficiency monitoring index, and perform weighted processing on the boiler efficiency monitoring index to obtain a comprehensive efficiency evaluation index;

[0054] Step 3: Based on the boiler efficiency monitoring indicators and boiler combustion history data, a data optimization model is established by the least squares method and the boiler efficiency is evaluated by using a clustering algorithm;

[0055] Step 4: Based on the boiler combustion history data, select the operation strategy with high frequency and confidence by using the association rule mining algorithm and setting the minimum support and minimum confidence parameters;

[0056] Step 5: Obtain the PINN neural network model based on the boiler combustion history data and the PINN neural network structure;

[0057] Step 6: Input the real-time boiler combustion data into the PINN neural network model to predict boiler efficiency.

[0058] The intelligent optimization is to establish an algorithm fitness function according to the optimization target, and then use the intelligent recommendation result as the initial value calculation of the GA algorithm, the non-dominated sorting genetic algorithm, the adaptive particle swarm algorithm and several swarm optimization algorithms and select the optimal value to obtain the optimal solution of the intelligent optimization; wherein the GA algorithm includes the following steps:

[0059] Step I: Initialize the population: Input the intelligent recommendation results into the GA algorithm population as the initial value, where the GA algorithm population size is 100. The population size will affect the results and efficiency of genetic optimization. If the population size is too small, the optimization performance will be reduced and it is easy to fall into the local optimal solution. If the population size is too large, the calculation complexity will increase.

[0060] Step II: Selection: Select individuals with higher fitness as parents according to the algorithm fitness function;

[0061] Step III: Crossover: Cross the parent generation to generate new offspring, with a crossover probability of 0.7;

[0062] Step IV: Mutation: The offspring are mutated with a mutation probability of 0.001;

[0063] Step V: Update the population: Use the parent and child generations to update the population;

[0064] Step VI: Iterate the above steps until the termination condition is met. In the present invention, the iteration is 1000 times;

[0065] Step VII: Decoding: Obtain the optimal variable value based on the optimal solution.

[0066] The safety check is performed by checking whether the variables are within the specified range and whether the constraints are met. The optimization results are safety checked to verify whether the safety boundaries and constraints are met.

[0067] The present invention uses intelligent recommendation, intelligent optimization and safety verification intelligent optimization methods to further improve boiler efficiency, enhance the accuracy of the boiler combustion system model, and make the intelligent optimization effect more obvious.

[0068] The above describes in detail the implementation modes of the present invention in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes. The implementation modes can still be changed within the knowledge scope of ordinary technicians in the field. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A boiler intelligent combustion optimization method based on a module-level mechanism and a mathematical hybrid model, which is divided into a boiler key indicator prediction model design and a boiler intelligent combustion optimization model design, and is characterized by: The method comprises the following steps, wherein steps 1) to 2) are for designing a boiler key indicator prediction model, and steps 3) to 6) are for designing a boiler intelligent combustion optimization model: 1) Based on the boiler system process and mechanism, the boiler system is divided into air preheating module, coal grinding module, coal quality analysis module, slag analysis module, boiler combustion module, economizer module, water wall module, superheater module, high pressure cylinder module, high pressure heating module, reheater module and medium pressure cylinder module; 2) Modeling the modules that cannot be modeled using a clear mechanism in step 1) using physical information fusion network technology; first build a mechanism model skeleton for the modules with clear mechanisms but unclear key parameters in step 1), and then use neural network technology to fit and correct the model coefficients in real time; Based on the boiler module model and the boiler operation history data, a prediction model for key parameters of the boiler is established through prediction model input feature screening, prediction model performance optimization and prediction model optimization; the prediction model input feature screening uses grey correlation and Spearman correlation coefficient algorithms to screen out several parameters with the highest correlation with the quantity to be predicted from a large number of parameters, and further screens parameters with physical significance based on manual experience, and uses the selected parameters as input feature quantities of the prediction model; further analyzes the correlation between the historical time series data of the quantity to be predicted and the current moment data, and selects the top N time series values ​​with the correlation value as the input of the time series prediction model; the prediction model performance optimization is optimized through model feature screening and variance analysis; the prediction model optimization is to select a model with the highest prediction accuracy from several prediction models of a certain key parameter as the prediction model to perform real-time prediction of the key parameter; the key parameters predicted by the prediction model are used as the input of the boiler intelligent combustion optimization model; 3) Use upper and lower threshold measures, mutation detection measures, multi-point anomaly detection measures, regression detection measures, energy conservation and physical property equation calibration measures to calibrate the real-time boiler data collected by various sensors as input data for the boiler intelligent combustion optimization model; 4) At the same time, the boiler combustion data that cannot be directly obtained is obtained through manual sampling and offline test data entry; 5) Design the optimization objectives and optimization decision variables of the boiler combustion optimization model; 6) Optimizing boiler combustion according to the optimization target and optimization decision variables of boiler combustion, wherein the boiler combustion optimization includes intelligent recommendation, intelligent optimization and safety verification; The intelligent recommendation is to establish a data optimization and clustering model based on boiler efficiency monitoring indicators, boiler combustion efficiency and historical data, and use physical information fusion network technology to build the framework of the intelligent recommendation model; The intelligent optimization is to establish an algorithm fitness function according to the optimization target, and then use the intelligent recommendation result as the initial value calculation of the genetic algorithm, the non-dominated sorting genetic algorithm, the adaptive particle swarm algorithm and several swarm optimization algorithms and select the optimal value to obtain the optimal solution of the intelligent optimization; The safety check is performed by checking whether the variables are within the specified range and whether the constraints are met. The optimization results are safety checked to verify whether the safety boundaries and constraints are met.

2. The method for optimizing boiler intelligent combustion based on module-level mechanism and mathematical hybrid model according to claim 1 is characterized in that: In the physical information fusion network technology in step 2), by establishing the objective function L = λ*Loss(y,y pred )+(1-λ)*║F(x)║ 2 Complete the modeling, where L is the overall objective function, Loss is the loss function of the neural network, y is the observed value, y pred is the predicted value of the neural network, F(x) is the residual of the physical equation, and λ is the parameter that weighs the loss function and the physical equation, which is set to 0.

5.

3. The boiler intelligent combustion optimization method based on module-level mechanism and mathematical hybrid model according to claim 1 is characterized by: The Spearman correlation coefficient method in step 2) is based on Calculate, where R i , S i is the i-th variable in the two sets of data after correlation sorting, is the average of the two sets of data, and N is the amount of data.

4. The boiler intelligent combustion optimization method based on module-level mechanism and mathematical hybrid model according to claim 1 is characterized by: The upper and lower threshold measures in step 3) are used to determine whether single-point data is abnormal; the mutation detection measure is based on the ARIMA time series prediction algorithm combined with the 3-sigma principle to determine data reliability; the multi-point anomaly detection measure is based on the PCA and AAKR signal reconstruction algorithm to detect the high-dimensional spatial distribution information of multi-sensor variables and local data anomaly information; the regression detection measure is based on LSTM regression detection for feature collection and preprocessing; the energy conservation and physical property equation calibration measure is used to calibrate information.

5. The method for optimizing boiler intelligent combustion based on module-level mechanism and mathematical hybrid model according to claim 4 is characterized in that: The specific steps of the mutation detection measure are: first, preprocess the historical data of boiler combustion, then establish a model based on the ARIMA time series prediction algorithm for the preprocessed data, and then use the established model to predict the latest boiler combustion data and use the 3-sigma principle to judge the data reliability. If the predicted value is not within the reliable data range, the data is judged to be mutation data; if the predicted value is within the reliable data range, the data is judged to be reliable data.

6. The boiler intelligent combustion optimization method based on module-level mechanism and mathematical hybrid model according to claim 1 is characterized by: The framework of building the intelligent recommendation model in step 6) includes the following steps: Step 1: Preprocess boiler combustion operation data through outlier detection, missing value filling and feature selection; Step 2: Determine the boiler efficiency monitoring index, and perform weighted processing on the boiler efficiency monitoring index to obtain a comprehensive efficiency evaluation index; Step 3: Based on the boiler efficiency monitoring indicators and boiler combustion history data, a data optimization model is established by the least squares method and the boiler efficiency is evaluated by using a clustering algorithm; Step 4: Based on the boiler combustion history data, select the operation strategy with high frequency and confidence by using the association rule mining algorithm and setting the minimum support and minimum confidence parameters; Step 5: Based on the boiler combustion history data and physical information fusion network technology, a physical information fusion network model is obtained; Step 6: Input the real-time boiler combustion data into the physical information fusion network model to predict boiler efficiency.

7. The method for optimizing boiler intelligent combustion based on module-level mechanism and mathematical hybrid model according to claim 1 is characterized in that: The genetic algorithm in step 6) comprises the following steps: Step I: Initialize the population: input the intelligent recommendation results into the genetic algorithm population as the initial value; Step II: Selection: Select individuals with higher fitness as parents according to the algorithm fitness function; Step III: Crossover: Cross the parent generation to generate new offspring; Step IV: Mutation: Mutate the offspring; Step V: Update the population: Use the parent and child generations to update the population; Step VI: Iterate the above steps until the termination condition is met; Step VII: Decoding: Obtain the optimal variable value based on the optimal solution.

8. The method for optimizing boiler intelligent combustion based on module-level mechanism and mathematical hybrid model according to claim 7 is characterized by: In the step I, the population size of the genetic algorithm is 100, in the step III, the crossover probability is 0.7, in the step IV, the mutation probability is 0.001, and in the step VI, the iteration is 1000 times.

Images