CN114186472A - Multi-input multi-output urban solid waste incineration process model design method - Google Patents

Multi-input multi-output urban solid waste incineration process model design method Download PDF

Info

Publication number
CN114186472A
CN114186472A CN202110998327.2A CN202110998327A CN114186472A CN 114186472 A CN114186472 A CN 114186472A CN 202110998327 A CN202110998327 A CN 202110998327A CN 114186472 A CN114186472 A CN 114186472A
Authority
CN
China
Prior art keywords
layer
input
output
solid waste
waste incineration
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110998327.2A
Other languages
Chinese (zh)
Other versions
CN114186472B (en
Inventor
乔俊飞
丁海旭
汤健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Technology
Original Assignee
Beijing University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Technology filed Critical Beijing University of Technology
Priority to CN202110998327.2A priority Critical patent/CN114186472B/en
Publication of CN114186472A publication Critical patent/CN114186472A/en
Application granted granted Critical
Publication of CN114186472B publication Critical patent/CN114186472B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Hardware Design (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Geometry (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Incineration Of Waste (AREA)

Abstract

Aiming at the problem that a controlled object model is difficult to establish in the urban solid waste incineration process, the invention designs a multi-input multi-output urban solid waste incineration process model design method; firstly, describing a core process flow of the urban solid waste incineration process and analyzing influence factors of a model; then, a modeling strategy of a controlled object model is designed according to the control characteristics of the urban solid waste incineration process, and the strategy consists of a working condition identification module, a data preprocessing module, a characteristic reduction module, a controlled object model training module and a controlled object model testing module; finally, the effectiveness of the controlled object model in the urban solid waste incineration process is shown through experiments, and a foundation is laid for researching an optimization control algorithm in the urban solid waste incineration process.

Description

Multi-input multi-output urban solid waste incineration process model design method
Technical Field
Aiming at the problems that the internal mechanism of the urban solid waste incineration process is complex and the controlled object model is difficult to establish, the invention designs a multi-input multi-output controlled object model based on a Takagi-Sugeno type fuzzy neural network, which consists of a working condition identification module, a data preprocessing module, a characteristic reduction module, a controlled object model training module and a controlled object model testing module, solves the problems that the internal mechanism of the urban solid waste incineration process is difficult to analyze, the multivariable coupling is strong and the internal rule is difficult to dig, and lays a model foundation for researching the optimal control of the urban solid waste incineration process.
Background
The urban solid waste incineration technology has the outstanding advantages of reduction, resource utilization, harmlessness and the like, and becomes one of the main technical means for treating urban solid waste in the world at present. In 2016, 303 solid waste incineration power plants operated in China had total treatment capacity of 3.04X 105t/day, wherein 220 solid waste incineration power plants using mechanical grate furnaces account for over 72%, the grate furnaces have become the main incinerator type used for solid waste incineration in Chinese cities. The urban solid waste incineration based on the grate furnace is a process with a plurality of uncertain characteristics such as strong nonlinearity, strong coupling, large time variation and the like, a complex process necessarily relates to the problems in a plurality of control fields, the solid waste incineration state can be stabilized and the operation efficiency can be improved only by means of an advanced control technology, and the establishment of an accurate controlled object model is the basis and necessary preparation for implementing the control technology. Therefore, the method has important significance for the design and research of the object model in the urban solid waste incineration process.
The traditional controlled object model is generally constructed based on mechanism analysis, and is called a mechanism model, also called a white box model. The mechanism model is constructed according to the principles of a material balance equation, an energy balance equation, a biological law, chemical kinetics and the like, and a relatively accurate mathematical model is established by deducing the functional relationship among the operating variables, the state variables and the controlled variables. The mechanism model has the capability of intuitively reflecting the relation between the internal rules and the structure of the system, however, different from the traditional complex industrial process, the raw material used in the urban solid waste incineration process is solid waste, and the urban solid waste incineration process has the characteristic of complexity and changeability essentially. The solid waste components are influenced by a plurality of factors, including seasonal climate, classification degree of solid waste, living standard, living habits and environmental awareness of people in the area and the like. For a strong nonlinear industrial process such as urban solid waste incineration, a model constructed based on mechanism analysis is difficult to analyze the property and the internal mechanism of a strong nonlinear system and is difficult to be suitable for the solid waste incineration process under multiple working conditions. In recent years, with the rise of artificial intelligence, a machine learning method based on data driving provides a solution for modeling a control object in the urban solid waste incineration process.
The data-driven model is constructed by mining the mapping relation of the input and output data of the system, and is also called as a black box model. The artificial neural network is widely applied to process analysis of a complex industrial system due to good learning ability, computing ability and nonlinear approximation ability, and has important application value when being applied to design of a controlled object model with unknown internal mechanism. Artificial neural networks have been increasingly used in modeling controlled objects in industrial processes, and have become a research hotspot in time.
The urban solid waste incineration process is a typical industrial process with multiple inputs and multiple outputs, the solid waste heat value is difficult to determine, the internal mechanical reaction is complex, the coupling of multiple operation quantities and controlled quantities is serious, the system rules are difficult to excavate, and the urban solid waste incineration process has typical fuzzy characteristics. Aiming at complex industrial processes such as urban solid waste incineration, the fuzzy neural network provides a good solution. The fuzzy neural network is taken as a fuzzy self-adaptive scheme, has the nonlinear processing and analyzing capability of a fuzzy system and the parameter learning and dynamic optimization capability of an artificial neural network, is widely researched in recent years, becomes an important branch in intelligent calculation and neuroscience, and is a technology superior to the technology for independent use of the artificial neural network and the fuzzy system.
According to the analysis, the invention designs a multi-input multi-output controlled object model based on the Takagi-Sugeno fuzzy neural network aiming at the characteristics of the urban solid waste incineration process. Firstly, identifying urban solid waste incineration operating conditions according to an incineration mechanism and expert experience and carrying out data pretreatment; then, extracting key operation quantity and controlled quantity capable of reflecting system state; then, a plurality of back-part sub-networks are constructed, local parameters and overall parameters of the network are optimized by adopting a gradient descent algorithm, and the convergence precision of the model and the output synchronism are ensured; and finally, verifying the validity of the controlled object model through the process data of a certain solid waste incineration plant in Beijing.
Disclosure of Invention
The invention obtains a multi-input multi-output controlled object model based on a Takagi-Sugeno type fuzzy neural network, the model consists of a working condition recognition module, a data preprocessing module, a characteristic reduction module, a controlled object model training module and a controlled object model testing module, thereby realizing the accurate prediction of key controlled variables, solving the problem that the controlled object model is difficult to establish in the urban solid waste incineration process and laying a foundation for researching the urban solid waste incineration optimization control;
the invention adopts the following technical scheme and implementation steps:
a design method of a multi-input multi-output urban solid waste incineration process model comprises the following steps:
1. a multi-input multi-output urban solid waste incineration process model design method is characterized by comprising the following steps:
(1) a working condition identification module: the module constructs a working condition identification expert evaluation mechanism based on primary air pressure, divides the working conditions according to primary air pressure set values and further constructs corresponding controlled object models according to different working conditions;
(2) a data preprocessing module: the module preprocesses acquired data through abnormal data elimination and data normalization, and comprises the following calculation steps:
removing abnormal data: firstly, detecting the normal distribution of data by drawing a quantile graph, then eliminating abnormal data by a 3 sigma criterion, and collecting key controlled variables at 1-T: the main steam flow, the furnace temperature and the flue gas oxygen content are defined as Ys(T), where s 1,2,., q, q is 3, T1, 2,.., T, calculating Ys(t) corresponding residual error εs(t) is:
Figure BDA0003234801030000031
calculating the standard deviation sigma of the data setsComprises the following steps:
Figure BDA0003234801030000032
when Y iss(t) corresponding residual error εs(t) when the following conditions are satisfied:
|εs(t)|＞3σs (3)
then to this Ys(T) performing a culling operation while letting T be T-1;
data normalization processing: extracting key operation variables in the urban solid waste incineration process: dry section grate air flow (left 1, right 1, left 2, right 2), combustion 1 section grate air flow (left 1, right 1, left 2, right 2), combustion 2 section furnace exhaust air flow (left 1, right 1, left 2, right 2), burn-out section grate air flow (left, right), overfire air flow, dry section grate velocity (left inside, right inside, left outside, right outside), combustion 1 section grate velocity (left inside, right inside, left outside, right outside), combustion 2 section grate velocity (left inside, right inside, left outside, right outside) and burn-out section grate velocity (left inside, right inside), defined as Xi(T), where i 1,2, N is 29, T1, 2, T, X, y, and y, yi(t) and Ys(t) carrying out normalization processing, wherein the calculation formula is as follows:
Figure BDA0003234801030000033
Figure BDA0003234801030000034
in the formula, xi(t) represents data Xi(t) normalized value, ys(t) represents data Ys(t) normalized value, XiAll data representing the ith parameter during the acquisition period, YsAll data representing the s-th parameter over the acquisition time period;
(3) a feature reduction module: calculate the above key manipulated variable xi(t) and the key controlled variable ysPearson correlation between (t)Coefficient, defining Pearson's correlation coefficient as rhodsThe calculation method comprises the following steps:
Figure BDA0003234801030000041
according to the calculation result, according to rhodsThe absolute values of (a) are sorted, the operation variables sorted to the top 3 are selected and marked as xi(t), wherein i ═ 1, 2.., n, n is 3;
(4) the multi-input multi-output Takagi-Sugeno type fuzzy neural network training module comprises: the model structure of this modular design comprises preceding network and back-part network two parts, and wherein preceding network includes that input layer, membership function layer, rule layer, back-part layer and output layer are 5 layers altogether, and back-part network includes that input layer, rule layer and back-part layer are 3 layers altogether, describes as follows to its mathematics:
inputting a layer: this layer has a total of n neurons, n being 3, whose role is to pass the input value, when the t-th sample comes in, the output of the input layer is:
xi(t),i＝1,2,...,n (7)
membership function layer: the layer has n multiplied by m neurons, m is 12, the output of each node represents the membership value of the corresponding input quantity, and the membership function is as follows:
Figure BDA0003234801030000042
in the formula, cij(t) and δij(t) is the center and width of the membership function, respectively, and the initial value is generated by a rand random function within the range of [0,2 ]]Random real numbers uniformly distributed among them;
third, rule layer: the layer is provided with m neurons, fuzzy connected multiplication operators are used as fuzzy logic rules, and the output of the rule layer is as follows:
Figure BDA0003234801030000043
fourthly, a back part layer: the layer has m × q neurons, q is 3, each node executes linear summation of T-S fuzzy rules, and the layer is used for calculating the back-piece parameters output by each rule correspondingly
Figure BDA0003234801030000051
The back-piece parameters are calculated by a back-piece network, n +1 variables are introduced into an input layer of the back-piece network, wherein the input of a 0 th node is a constant, namely x0And (t) 1, transmitting the back-part parameters back to a back-part layer of the front-part network, wherein the calculation process is as follows:
Figure BDA0003234801030000052
in the formula (I), the compound is shown in the specification,
Figure BDA0003234801030000053
the initial value of the parameter is set to 0.3, x for fuzzy system0(t),x1(t),…, xn(t) is an input variable;
outputting a layer: the layer is provided with q output nodes, each node performs weighted summation on input parameters, and the calculation formula is as follows:
Figure BDA0003234801030000054
model parameter learning: using a gradient descent algorithm to adjust network parameters, first, an error calculation method is defined as follows:
Figure BDA0003234801030000055
in the formula, ys(t) is the s-th actual output corresponding to the t-th input sample,
Figure BDA0003234801030000056
is the s-th calculated output corresponding to the t-th input sample, es(t) isThe error between the two is defined as follows according to the error to the center, width and fuzzy system parameter updating algorithm of the network:
Figure BDA0003234801030000057
Figure BDA0003234801030000058
Figure BDA0003234801030000059
wherein eta is online learning rate, and eta has a value range of [0.01,0,05 ]]，cij(t-1)、δij(t-1) and
Figure BDA00032348010300000510
respectively inputting the center and width of a network membership function layer and the parameters of a fuzzy system when the t-1 th sample is input, and inputting training sample data x after the parameters are updatedi(T +1), repeating the steps from (i) to (ii) until all training samples are input, the number of the training samples is 80% of the total number of samples T, and then performing iterative training on the model until the iteration number reaches the maximum iteration value Itmax，ItmaxIs 500;
(5) after the model training is finished, the building of the multi-input multi-output urban solid waste incineration process model is finished, at the moment, the primary air flow, the secondary air flow and the grate speed are input into the model, and then the model outputs the main steam flow, the hearth temperature and the smoke oxygen content.
The invention is mainly characterized in that:
(1) the method solves the problems that the internal mechanism of the urban solid waste incineration process is difficult to analyze, the multivariable coupling is strong, and the internal rule is difficult to excavate, and lays a model foundation for researching the optimization control of the urban solid waste incineration process;
(2) aiming at the process characteristics of domestic urban solid waste incineration, the modeling strategy with multi-working-condition identification and characteristic reduction is designed, and the model has better robustness and applicability;
(3) the controlled object model established by the invention has multi-output learning capability, accurately fits a plurality of controlled quantities at the same time by utilizing the complementary information among multiple tasks, and updates network parameters on line;
(4) the modeling method for the controlled object in the urban solid waste incineration process based on the Takagi-Sugeno type fuzzy neural network multi-input multi-output controlled object model has the advantages of strong learning capability, high modeling precision and wide application value;
drawings
FIG. 1 is a process flow diagram of the municipal solid waste incineration process of the invention
FIG. 2 is a control flow chart of the municipal solid waste incineration process of the invention
FIG. 3 is a controlled object model construction method based on data-driven urban solid waste incineration process of the invention
FIG. 4 is a multi-input multi-output model of the present invention based on Takagi-Sugeno type fuzzy neural network
FIG. 5 is a diagram of the model training process RMSE variation of the present invention
FIG. 6 is a model training sample fitting effect of the present invention
FIG. 7 is a sample fit effect of the model test of the present invention
FIG. 8 is a graph of the fitting effect of the furnace temperature training samples of the present invention
FIG. 9 is a graph of the fitting effect of the smoke oxygen content training sample of the present invention
FIG. 10 is a graph of the effect of the fit of the main steam flow test sample of the present invention
FIG. 11 is a graph of the fitting effect of the furnace temperature test samples of the present invention
FIG. 12 is a graph showing the fitting effect of the smoke oxygen content test sample according to the present invention
FIG. 13 is a test error plot of a main steam flow test sample of the present invention
FIG. 14 is a test error plot of a furnace temperature test sample of the present invention
FIG. 15 is a graph showing the test error of the sample for measuring the oxygen content in flue gas according to the present invention
Detailed Description
The invention obtains a multi-input multi-output controlled object model based on a Takagi-Sugeno type fuzzy neural network, the model consists of a working condition recognition module, a data preprocessing module, a characteristic reduction module, a controlled object model training module and a controlled object model testing module, thereby realizing the accurate prediction of key controlled variables, solving the problem that the controlled object model is difficult to establish in the urban solid waste incineration process and laying a foundation for researching the urban solid waste incineration optimization control;
the experiment collects the process data of a certain urban solid waste incineration power plant, the sampling frequency is 1 s/time, and 2 multiplied by 10 is obtained through collection5Group data were used as experimental samples;
a design method of a multi-input multi-output urban solid waste incineration process model comprises the following steps:
(1) a working condition identification module: the module constructs a working condition identification expert evaluation mechanism based on primary air pressure, divides the working conditions according to primary air pressure set values and further constructs corresponding controlled object models according to different working conditions;
(2) a data preprocessing module: the module preprocesses acquired data through abnormal data elimination and data normalization, and comprises the following calculation steps:
removing abnormal data: firstly, detecting the normal distribution of data by drawing a quantile graph, then eliminating abnormal data by a 3 sigma criterion, and collecting key controlled variables at 1-T: the main steam flow, the furnace temperature and the flue gas oxygen content are defined as Ys(T), where s 1,2,., q, q is 3, T1, 2,.., T, calculating Ys(t) corresponding residual error εs(t) is:
Figure BDA0003234801030000071
calculating the standard deviation sigma of the data setsComprises the following steps:
Figure BDA0003234801030000072
when Y iss(t) corresponding residual error εs(t) when the following conditions are satisfied:
|εs(t)|＞3σs (3)
then to this Ys(T) performing a culling operation while letting T be T-1;
data normalization processing: extracting key operation variables in the urban solid waste incineration process: dry section grate air flow (left 1, right 1, left 2, right 2), combustion 1 section grate air flow (left 1, right 1, left 2, right 2), combustion 2 section furnace exhaust air flow (left 1, right 1, left 2, right 2), burn-out section grate air flow (left, right), overfire air flow, dry section grate velocity (left inside, right inside, left outside, right outside), combustion 1 section grate velocity (left inside, right inside, left outside, right outside), combustion 2 section grate velocity (left inside, right inside, left outside, right outside) and burn-out section grate velocity (left inside, right inside), defined as Xi(T), where i 1,2, N is 29, T1, 2, T, X, y, and y, yi(t) and Ys(t) carrying out normalization processing, wherein the calculation formula is as follows:
Figure BDA0003234801030000081
Figure BDA0003234801030000082
in the formula, xi(t) represents data Xi(t) normalized value, ys(t) represents data Ys(t) normalized value, XiAll data representing the ith parameter during the acquisition period, YsAll data representing the s-th parameter over the acquisition time period;
(3) a feature reduction module: calculate the above key manipulated variable xi(t) and the key controlled variable ysBetween (t)Pearson correlation coefficient, defined as ρdsThe calculation method comprises the following steps:
Figure BDA0003234801030000083
according to the calculation result, according to rhodsThe absolute values of (a) are sorted, the operation variables sorted to the top 3 are selected and marked as xi(t), wherein i ═ 1, 2.., n, n is 3;
(4) the multi-input multi-output Takagi-Sugeno type fuzzy neural network training module comprises: the model structure of this modular design comprises preceding network and back-part network two parts, and wherein preceding network includes that input layer, membership function layer, rule layer, back-part layer and output layer are 5 layers altogether, and back-part network includes that input layer, rule layer and back-part layer are 3 layers altogether, describes as follows to its mathematics:
inputting a layer: this layer has a total of n neurons, n being 3, whose role is to pass the input value, when the t-th sample comes in, the output of the input layer is:
xi(t),i＝1,2,...,n (7)
membership function layer: the layer has n multiplied by m neurons, m is 12, the output of each node represents the membership value of the corresponding input quantity, and the membership function is as follows:
Figure BDA0003234801030000084
in the formula, cij(t) and δij(t) is the center and width of the membership function, respectively, and the initial value is generated by a rand random function within the range of [0,2 ]]Random real numbers uniformly distributed among them;
third, rule layer: the layer is provided with m neurons, fuzzy connected multiplication operators are used as fuzzy logic rules, and the output of the rule layer is as follows:
Figure BDA0003234801030000091
fourthly, a back part layer: the layer has m × q neurons, q is 3, each node executes linear summation of T-S fuzzy rules, and the layer is used for calculating the back-piece parameters output by each rule correspondingly
Figure BDA0003234801030000092
The back-piece parameters are calculated by a back-piece network, n +1 variables are introduced into an input layer of the back-piece network, wherein the input of a 0 th node is a constant, namely x0And (t) 1, transmitting the back-part parameters back to a back-part layer of the front-part network, wherein the calculation process is as follows:
Figure BDA0003234801030000093
in the formula (I), the compound is shown in the specification,
Figure BDA0003234801030000094
the initial value of the parameter is set to 0.3, x for fuzzy system0(t),x1(t),…, xn(t) is an input variable;
outputting a layer: the layer is provided with q output nodes, each node performs weighted summation on input parameters, and the calculation formula is as follows:
Figure BDA0003234801030000095
model parameter learning: using a gradient descent algorithm to adjust network parameters, first, an error calculation method is defined as follows:
Figure BDA0003234801030000096
in the formula, ys(t) is the s-th actual output corresponding to the t-th input sample,
Figure BDA0003234801030000097
is the s-th calculated output corresponding to the t-th input sample, es(t) as the error between the two, the algorithm for updating the center, width and fuzzy system parameters of the network according to the error is defined as follows:
Figure BDA0003234801030000098
Figure BDA0003234801030000099
Figure BDA0003234801030000101
wherein eta is online learning rate, and eta has a value range of [0.01,0,05 ]]，cij(t-1)、δij(t-1) and
Figure BDA0003234801030000102
respectively inputting the center and width of a network membership function layer and the parameters of a fuzzy system when the t-1 th sample is input, and inputting training sample data x after the parameters are updatedi(T +1), repeating the steps from (i) to (ii) until all training samples are input, the number of the training samples is 80% of the total number of samples T, and then performing iterative training on the model until the iteration number reaches the maximum iteration value Itmax，ItmaxIs 500;
(5) after the model training is finished, the building of the multi-input multi-output urban solid waste incineration process model is finished, at the moment, the primary air flow, the secondary air flow and the grate speed are input into the model, and then the model outputs the main steam flow, the hearth temperature and the smoke oxygen content.
FIG. 1 is a process flow diagram of the municipal solid waste incineration process of the invention
FIG. 2 is a control flow chart of the municipal solid waste incineration process of the invention
FIG. 3 is a diagram of a multi-input multi-output model structure based on a Takagi-Sugeno type fuzzy neural network of the present invention
FIG. 4 is a graph of the main steam flow training process RMSE variation of the present invention
FIG. 5 is a graph of RMSE variation in the furnace temperature training process of the present invention
FIG. 6 is a graph of RMSE variation in the smoke oxygen content training process of the present invention
FIG. 7 is a graph of the fitting effect of the main steam flow training samples of the present invention
FIG. 8 is a graph of the fitting effect of the furnace temperature training samples of the present invention
FIG. 9 is a graph of the fitting effect of the smoke oxygen content training sample of the present invention
FIG. 10 is a graph of the effect of the fit of the main steam flow test sample of the present invention
FIG. 11 is a graph of the fitting effect of the furnace temperature test samples of the present invention
FIG. 12 is a graph showing the fitting effect of the smoke oxygen content test sample according to the present invention
FIG. 13 is a test error plot of a main steam flow test sample of the present invention
FIG. 14 is a test error plot of a furnace temperature test sample of the present invention
FIG. 15 is a graph showing the test error of the sample for measuring the oxygen content in flue gas according to the present invention
The process flow of the municipal solid waste incineration process is shown in figure 1; the control flow of the municipal solid waste incineration process is shown in fig. 2; the structure of a multi-input multi-output model based on a Takagi-Sugeno type fuzzy neural network is shown in FIG. 3; the training sample data is used as the input of the model, and the RMSE change in the model training process is shown in FIGS. 4-6; fig. 4 shows the change in the primary steam flow training process RMSE, X-axis: training step number, Y-axis: training the RMSE value; FIG. 5 shows the furnace temperature training process RMSE variation, X-axis: training step number, Y-axis: training the RMSE value; FIG. 6 shows the RMSE variation during the smoke oxygen content training, X-axis: training step number, Y-axis: training the RMSE value; the fitting effect of the model training samples is shown in fig. 7-9; fig. 7 is the main steam flow training sample fitting effect, X-axis: sample number, Y-axis: the unit of the main steam flow is t/h, a black line is predicted output, and a gray line is actual output; fig. 8 is a furnace temperature training sample fitting effect, X-axis: sample number, Y-axis: the unit of the hearth temperature is that the black line is the predicted output and the gray line is the actual output; fig. 9 is a smoke oxygen content training sample fitting effect, X-axis: sample number, Y-axis: the oxygen content of the smoke is calculated in unit of percent, a black line is used as the predicted output, and a gray line is used as the actual output; inputting test sample data as a trained model, wherein the fitting effect of the model test sample is shown in fig. 10-12; fig. 10 is the main steam flow test sample fitting effect, X-axis: sample number, Y-axis: the unit of the main steam flow is t/h, a black line is used for prediction output, and a gray line is used for actual output; fig. 11 is a furnace temperature test sample fitting effect, X-axis: sample number, Y-axis: the unit of the hearth temperature is that black lines are used as predicted output and gray lines are used as actual output; fig. 12 is a sample fitting effect of the flue gas oxygen content test, X-axis: sample number, Y-axis: the oxygen content of the smoke, the unit is%, the black line is the predicted output, and the gray line is the actual output; the test error of the model test sample is shown in fig. 13 to 15; fig. 13 is the main steam flow test sample test error, X-axis: sample number, Y-axis: the unit of the main steam flow is t/h; FIG. 14 is a test error of a furnace temperature test sample, X-axis: sample number, Y-axis: the temperature of the hearth is measured in units of; FIG. 15 shows the test error of the sample for the oxygen content test of flue gas, X-axis: sample number, Y-axis: oxygen content of flue gas, unit is%; the result shows the effectiveness of the model in modeling the controlled object in the urban solid waste incineration process.

Claims (1)

1. A multi-input multi-output urban solid waste incineration process model design method is characterized by comprising the following steps:
(1) a working condition identification module: dividing the working conditions according to the primary air pressure set value, and further constructing corresponding controlled object models according to different working conditions;
(2) a data preprocessing module: the module preprocesses acquired data through abnormal data elimination and data normalization, and comprises the following calculation steps:
removing abnormal data: firstly, detecting the normal distribution of data by drawing a quantile graph, then eliminating abnormal data by a 3 sigma criterion, and collecting key controlled variables at 1-T: main steaming deviceThe steam flow, the hearth temperature and the oxygen content of the flue gas are defined as Ys(T), where s 1,2,., q, q is 3, T1, 2,.., T, calculating Ys(t) corresponding residual error εs(t) is:
Figure FDA0003234801020000011
calculating the standard deviation sigma of the data setsComprises the following steps:
Figure FDA0003234801020000012
when Y iss(t) corresponding residual error εs(t) when the following conditions are satisfied:
|εs(t)|＞3σs (3)
then to this Ys(T) performing a culling operation while letting T be T-1;
data normalization processing: extracting key operation variables in the urban solid waste incineration process: the air flow of the grate at the drying section (left 1, right 1, left 2, right 2), the air flow of the grate at the combustion 1 section (left 1, right 1, left 2, right 2), the air flow of the grate at the combustion 2 section (left 1, right 1, left 2, right 2), the air flow of the grate at the burn-out section (left, right), the secondary air flow, the grate speed at the drying section (left inner, right inner, left outer, right outer), the grate speed at the combustion 1 section (left inner, right inner, left outer, right outer), and the grate speed at the burn-out section (left inner, right inner) are defined as Xi(T), where i 1,2, N is 29, T1, 2, T, X, y, and y, yi(t) and Ys(t) carrying out normalization processing, wherein the calculation formula is as follows:
Figure FDA0003234801020000013
Figure FDA0003234801020000014
in the formula, xi(t) represents data Xi(t) normalized value, ys(t) represents data Ys(t) normalized value, XiAll data representing the ith parameter during the acquisition period, YsAll data representing the s-th parameter over the acquisition time period;
(3) a feature reduction module: calculate the above key manipulated variable xi(t) and the key controlled variable ys(t) Pearson's correlation coefficient, defined as ρdsThe calculation method comprises the following steps:
Figure FDA0003234801020000021
according to the calculation result, according to rhodsThe absolute values of (a) are sorted, the operation variables sorted to the top 3 are selected and marked as xi(t), wherein i ═ 1, 2.., n, n is 3;
(4) the multi-input multi-output Takagi-Sugeno type fuzzy neural network training module comprises: the model structure of this modular design comprises preceding network and back-part network two parts, and wherein preceding network includes that input layer, membership function layer, rule layer, back-part layer and output layer are 5 layers altogether, and back-part network includes that input layer, rule layer and back-part layer are 3 layers altogether, describes as follows to its mathematics:
inputting a layer: this layer has a total of n neurons, n being 3, whose role is to pass the input value, when the t-th sample comes in, the output of the input layer is:
xi(t),i＝1,2,...,n (7)
membership function layer: the layer has n multiplied by m neurons, m is 12, the output of each node represents the membership value of the corresponding input quantity, and the membership function is as follows:
Figure FDA0003234801020000022
in the formula, cij(t) and δij(t) is the center and width of the membership function, respectively, and the initial value is generated by a rand random function within the range of [0,2 ]]Random real numbers uniformly distributed among them;
third, rule layer: the layer is provided with m neurons, fuzzy connected multiplication operators are used as fuzzy logic rules, and the output of the rule layer is as follows:
Figure FDA0003234801020000023
fourthly, a back part layer: the layer has m × q neurons, q is 3, each node performs linear summation of T-S fuzzy rules, and the layer is used for calculating the back-piece parameters of the output corresponding to each rule
Figure FDA0003234801020000031
The back-piece parameters are calculated by a back-piece network, n +1 variables are introduced into an input layer of the back-piece network, wherein the input of a 0 th node is a constant, namely x0And (t) 1, transmitting the back-part parameters back to a back-part layer of the front-part network, wherein the calculation process is as follows:
Figure FDA0003234801020000032
in the formula (I), the compound is shown in the specification,
Figure FDA0003234801020000033
the initial value of the parameter is set to 0.3, x for fuzzy system0(t),x1(t),…,xn(t) is an input variable;
outputting a layer: the layer is provided with q output nodes, each node performs weighted summation on input parameters, and the calculation formula is as follows:
Figure FDA0003234801020000034
model parameter learning: using a gradient descent algorithm to adjust network parameters, first, an error calculation method is defined as follows:
Figure FDA0003234801020000035
in the formula, ys(t) is the s-th actual output corresponding to the t-th input sample,
Figure FDA0003234801020000036
is the s-th calculated output corresponding to the t-th input sample, es(t) as the error between the two, the algorithm for updating the center, width and fuzzy system parameters of the network according to the error is defined as follows:
Figure FDA0003234801020000037
Figure FDA0003234801020000038
Figure FDA0003234801020000039
wherein eta is online learning rate, and eta has a value range of [0.01,0,05 ]]，cij(t-1)、δij(t-1) and
Figure FDA00032348010200000310
inputting the center and width of a network membership function layer and the parameters of a fuzzy system when the t-1 th sample is input respectively, and inputting training sample data x after the parameters are updatedi(T +1), repeating the steps from (I) to (VI) until all training samples are input, the number of the training samples is 80% of the total number of samples T, and then carrying out iterative training on the model until the number of iterations reachesTo the maximum iteration value Itmax，ItmaxIs 500;
(5) after the model training is finished, the building of the multi-input multi-output urban solid waste incineration process model is finished, at the moment, the primary air flow, the secondary air flow and the grate speed are input into the model, and then the model outputs the main steam flow, the hearth temperature and the smoke oxygen content.
CN202110998327.2A 2021-08-27 2021-08-27 Design method of multi-input multi-output urban solid waste incineration process model Active CN114186472B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110998327.2A CN114186472B (en) 2021-08-27 2021-08-27 Design method of multi-input multi-output urban solid waste incineration process model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110998327.2A CN114186472B (en) 2021-08-27 2021-08-27 Design method of multi-input multi-output urban solid waste incineration process model

Publications (2)

Publication Number Publication Date
CN114186472A true CN114186472A (en) 2022-03-15
CN114186472B CN114186472B (en) 2023-05-02

Family

ID=80601030

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110998327.2A Active CN114186472B (en) 2021-08-27 2021-08-27 Design method of multi-input multi-output urban solid waste incineration process model

Country Status (1)

Country Link
CN (1) CN114186472B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115542740A (en) * 2022-10-07 2022-12-30 北京工业大学 Multivariable active disturbance rejection control method for urban solid waste incineration process based on data driving
CN115755779A (en) * 2022-08-27 2023-03-07 北京工业大学 Control-oriented data-driven urban solid waste incineration full-process modeling method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101476731A (en) * 2008-12-31 2009-07-08 华南理工大学 Refuse incineration control method based on refuse thermal value soft measurement
CN105864797A (en) * 2016-04-01 2016-08-17 浙江大学 Real-time prediction system and method for boiler entering heat value of circulating fluidized bed household garbage incineration boiler
CN112733876A (en) * 2020-10-28 2021-04-30 北京工业大学 Soft measurement method for nitrogen oxides in urban solid waste incineration process based on modular neural network
CN113191078A (en) * 2021-04-26 2021-07-30 北京工业大学 Method for forecasting flue gas temperature of primary combustion chamber in municipal solid waste incineration process

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101476731A (en) * 2008-12-31 2009-07-08 华南理工大学 Refuse incineration control method based on refuse thermal value soft measurement
CN105864797A (en) * 2016-04-01 2016-08-17 浙江大学 Real-time prediction system and method for boiler entering heat value of circulating fluidized bed household garbage incineration boiler
CN112733876A (en) * 2020-10-28 2021-04-30 北京工业大学 Soft measurement method for nitrogen oxides in urban solid waste incineration process based on modular neural network
CN113191078A (en) * 2021-04-26 2021-07-30 北京工业大学 Method for forecasting flue gas temperature of primary combustion chamber in municipal solid waste incineration process

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115755779A (en) * 2022-08-27 2023-03-07 北京工业大学 Control-oriented data-driven urban solid waste incineration full-process modeling method
CN115542740A (en) * 2022-10-07 2022-12-30 北京工业大学 Multivariable active disturbance rejection control method for urban solid waste incineration process based on data driving
CN115542740B (en) * 2022-10-07 2024-07-16 北京工业大学 Multivariable active disturbance rejection control method based on data driving for urban solid waste incineration process

Also Published As

Publication number Publication date
CN114186472B (en) 2023-05-02

Similar Documents

Publication Publication Date Title
CN114186472A (en) Multi-input multi-output urban solid waste incineration process model design method
Peña et al. Soft-computing models for soot-blowing optimization in coal-fired utility boilers
Elfaki et al. Prediction of electrical output power of combined cycle power plant using regression ANN model
Hou et al. Coordinated control system modeling of ultra-supercritical unit based on a new fuzzy neural network
CN114282443B (en) Residual service life prediction method based on MLP-LSTM supervised joint model
CN108224446A (en) A kind of automatic combustion Study on Decision-making Method for Optimization of Refuse Incineration Process
Yang et al. Forecasting of iron ore sintering quality index: A latent variable method with deep inner structure
Ding et al. MIMO modeling and multi-loop control based on neural network for municipal solid waste incineration
CN114527646A (en) Multi-loop quasi-diagonal recurrent neural network PID control method for urban solid waste incineration process
CN116451567A (en) Leakage assessment and intelligent disposal method for gas negative pressure extraction pipeline
Jin et al. Simplification of ANFIS based on importance-confidence-similarity measures
Yang et al. A novel exhaust gas temperature prediction method of hot blast stove
CN113887807B (en) Robot game tactics prediction method based on machine learning and evidence theory
CN113191078B (en) Flue gas temperature forecasting method for primary combustion chamber in urban household garbage incineration process
Gholamian et al. A hybrid intelligent system for multiobjective decision making problems
Li et al. Least Square Parallel Extreme Learning Machine for Modeling NO x Emission of a 300MW Circulating Fluidized Bed Boiler
CN115526433A (en) Power plant reheat flue gas baffle operation prediction method based on integrated hybrid model
Tarik et al. Optimization of neural network hyperparameters for gas turbine modelling using Bayesian optimization
Wang et al. Soft Sensor Modeling of Self-Organizing Interval Type-2 Fuzzy Neural Network Based on Adaptive Quantum-Behaved Particle Swarm Optimization Algorithm
Gao et al. Fuzzy chance-constrained programming for capital budgeting problem with fuzzy decisions
CN115542740B (en) Multivariable active disturbance rejection control method based on data driving for urban solid waste incineration process
Sedano et al. Learning and training techniques in fuzzy control for energy efficiency in buildings
CN113779858A (en) Combustion optimization method and system, storage medium and electronic equipment
Mozaffari et al. A robust time delay auto-regressive exogenous fuzzy inference system for real-time estimation of catalyst temperature over engines coldstart operation: a multiobjective implementation scenario
Ding et al. Multi-input and multi-output modeling method based on TS fuzzy neural network and its application

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant