A temperature prediction method for incinerator based on numerical simulation and interval type-2 fuzzy neural network

Abstract

The application discloses a kind of incinerator temperature prediction methods based on numerical simulation and interval two type fuzzy neural network, belong to urban solid waste incineration processing and industrial process intelligent modeling technical field.The method includes: obtaining incinerator initial operating parameter, establishes incinerator numerical simulation model;Orthogonal experimental design is carried out to controllable operating parameter in numerical simulation model, according to the result of orthogonal experimental design, runs numerical simulation model, obtains simulation data set under multiple working conditions;Interval two type fuzzy neural network is constructed, and it is trained using simulation data set, and temperature proxy model is obtained;Target operating parameter is input into temperature proxy model, and the temperature prediction value of incinerator outlet position is output.The application can quickly and accurately obtain the temperature of incinerator outlet without relying on physical sensor and without performing time-consuming numerical simulation calculation, provide reliable technical support for incinerator digital twin system construction and operation optimization control.

Description

Technical Field

[0001] This invention belongs to the field of intelligent modeling technology for urban solid waste incineration and industrial processes, and particularly relates to a method for predicting incinerator temperature based on numerical simulation and interval type II fuzzy neural network. Background Technology

[0002] MSWI incineration is one of the core disposal methods for achieving waste reduction, harmlessness, and resource recovery both domestically and internationally. In the daily operation of an MSWI plant, the average flue gas temperature at the outlet of the third flue of the incinerator is a crucial process performance indicator. This temperature value not only directly reflects the stability and completeness of combustion conditions within the furnace but also serves as the core basis for guiding operators to adjust key control parameters such as incineration rate, grate speed, and air distribution scheme. If the temperature is too low, it indicates incomplete combustion, which may lead to the generation and emission of harmful substances such as dioxins exceeding standards; if the temperature is too high, it may cause equipment safety problems such as furnace coking and superheater overheating. Therefore, accurate and rapid acquisition of the incinerator outlet temperature is a key prerequisite for ensuring the safe, environmentally friendly, and efficient operation of the incineration power plant.

[0003] Currently, the technical solutions for acquiring temperature information from MSWI incinerators can be mainly divided into the following two categories: (a) Direct measurement methods based on physical sensors; In engineering practice, most MSWI power plants use thermocouples or resistance temperature sensors installed at the outlet of the third flue of the incinerator for real-time temperature acquisition, and transmit the signals to the central control room for display and recording via a distributed control system (DCS). This solution is currently the most mainstream method for temperature acquisition in industrial sites.

[0004] However, the technical solution has the following serious defects in practical application: (1) The flue gas environment in the third flue of the incinerator is extremely harsh. The flue gas temperature is maintained in the high temperature range of 900 K to 1200 K for a long time. The flue gas is also mixed with high concentrations of dust particles, acidic corrosive gases (such as HCl and SO2) and alkaline metal vapors. The sensor probe is exposed to this environment for a long time and is very prone to high temperature oxidation, chemical corrosion and dust erosion wear, which leads to zero drift, sensitivity decay and even complete failure of the measuring element. Its reliable operating life is usually only a few months and the maintenance and replacement cost is high. (2) The sensor is a single-point measurement method. Its output value only represents the local temperature of the probe location. The temperature field of the third flue outlet section has a significant non-uniform distribution. Single-point measurement is difficult to fully reflect the average temperature level and temperature distribution characteristics of the section. (3) Due to the structural design of the incinerator body and the safety regulations of high temperature and high pressure, the installation position and insertion depth of the temperature sensor are strictly limited. The positions that can be used to set up measuring points are very limited. It is difficult to make up for the lack of representativeness of point measurement by adding sensors.

[0005] (ii) Numerical simulation methods based on computational fluid dynamics; To overcome the limitations of physical sensor measurement methods in terms of information dimension and spatial coverage, computational fluid dynamics numerical simulation technology has been gradually introduced into academia and engineering in recent years. This technology utilizes specialized simulation software to perform coupled numerical simulations of the combustion process on the solid fuel layer in the incinerator and the gas-phase turbulent combustion process inside the furnace. It can obtain complete temperature field distribution and detailed information on the concentration fields of flue gas components (O2, CO2, CO, NOx, etc.) in three-dimensional space, thus overcoming the spatial limitations of physical sensor point measurements.

[0006] Specifically, this numerical simulation method requires solving multiple partial differential equations simultaneously. These equations constitute a strongly coupled set of nonlinear partial differential equations, which need to be solved iteratively on a spatial grid using numerical discretization methods such as the finite volume method. A single simulation typically involves hundreds of thousands to millions of grid cells, with computation times lasting several hours or even tens of hours, and requires the hardware support of a high-performance computing server.

[0007] Although this technical solution has a significant advantage in terms of information richness, it also has the same prominent defects: (1) The single simulation calculation takes a long time, which cannot meet the hard requirements of digital twin systems for real-time data refresh (usually requiring a response time of seconds or minutes). (2) When it is necessary to conduct simulation analysis on different operating conditions (such as changes in incineration volume, adjustments to air distribution schemes, fluctuations in waste composition, etc.), the complete simulation process must be executed repeatedly, which increases the time cost sharply and seriously restricts its application in dynamic scenarios such as process optimization control and online decision support. (3) The high consumption of computing resources means that it is difficult to equip ordinary industrial sites with the corresponding computing infrastructure, which limits the engineering promotion of this technology.

[0008] In summary, the two mainstream methods for obtaining the outlet temperature of existing MSWI incinerators each have significant technical shortcomings: the physical sensor approach is limited by harsh environments, resulting in poor reliability and sparse measurement information that is difficult to comprehensively represent; the numerical simulation approach is limited by excessive computation time, failing to meet real-time requirements. These dilemmas constitute the core technical bottleneck for the practical application of MSWI digital twin technology, necessitating the development of a novel rapid temperature prediction method that balances prediction accuracy and real-time response capabilities to support digital twin-driven operational optimization and intelligent decision-making. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a method for predicting incinerator temperature based on numerical simulation and interval type-II fuzzy neural networks, comprising: Obtain the initial operating parameters of the incinerator, and establish a numerical simulation model of the incinerator based on the initial operating parameters; Orthogonal experimental design is performed on the controllable operating parameters in the numerical simulation model, and the numerical simulation model is run according to the orthogonal experimental design results to obtain simulation datasets under multiple working conditions. An interval type II fuzzy neural network is constructed, and the interval type II fuzzy neural network is trained using the simulation dataset to obtain a temperature proxy model; The target operating parameters are input into the temperature proxy model, and the temperature proxy model outputs the predicted temperature value at the incinerator outlet.

[0010] Optionally, obtaining the initial operating parameters of the incinerator and establishing a numerical simulation model of the incinerator based on the initial operating parameters includes: Obtain actual operating data of the incinerator, including primary air temperature, primary air flow rate, grate speed, and combustion rate; Based on the actual operating data, a numerical simulation model of the incinerator was established by coupling solid-phase combustion simulation software and gas-phase combustion simulation software. The actual operating data is used as boundary conditions to input the numerical simulation model, and the simulation output results are compared and verified with the field measured values ​​to obtain a reliable numerical simulation model.

[0011] Optionally, orthogonal experimental design is performed on the controllable operating parameters in the numerical simulation model, and the numerical simulation model is run according to the orthogonal experimental design results to obtain a simulation dataset under multiple operating conditions, including: Multiple controllable operating parameters that affect the combustion effect are selected from the incinerator's operating conditions, and multiple level values ​​are determined for each controllable operating parameter. An orthogonal experimental design method is adopted to generate an orthogonal experimental scheme based on the multiple controllable operating parameters and multiple level values ​​of each controllable operating parameter. The orthogonal experimental scheme includes multiple experimental conditions. The numerical simulation model was run sequentially according to each experimental condition in the orthogonal experimental scheme, and the temperature data at the incinerator outlet was recorded under each condition. The temperature data is preprocessed to obtain a simulation dataset.

[0012] Optionally, the preprocessing includes outlier removal, normalization, and dividing the dataset into training, validation, and test sets.

[0013] Optionally, the interval type II fuzzy neural network includes an input layer, a membership function layer, a rule layer, a consequent layer, and an output layer.

[0014] Optionally, the functions of each layer of the interval type II fuzzy neural network are as follows: The input layer receives input variables and transmits the input variables to the membership function layer; The membership function layer uses the interval type II Gaussian membership function to fuzzify the input variables into interval-form membership values; The rule layer calculates the upper and lower emission intensities of each fuzzy rule based on the membership value; The backing layer performs type reduction based on the upper and lower emission intensities to obtain upper and lower output values; The output layer deblurs the upper and lower output values ​​and outputs the predicted temperature value at the incinerator outlet.

[0015] Optionally, the interval type II Gaussian membership function introduces an adaptive shape factor, which is a positive parameter. This adaptive shape factor, together with the center and width of the interval type II Gaussian membership function, serves as a network parameter and is optimized and adjusted through a hierarchical learning algorithm.

[0016] Optionally, the consequent layer is type-reduction and deblurring performed using the Begian-Melek-Mendel method.

[0017] Optionally, a hierarchical learning algorithm is used to update the parameters of the interval type II fuzzy neural network, wherein the least squares estimation method is used to learn the linear prior parameters and shape factor, and the gradient descent method is used to learn the nonlinear result parameters.

[0018] Optionally, the input variables of the temperature proxy model are controllable operating parameters of the operating conditions, including combustion rate, grate speed, inlet air temperature of the drying grate, inlet air temperature of the combustion grate, outlet air temperature of the primary air heater, primary air flow rate of the drying grate, primary air flow rate of combustion grate 1, primary air flow rate of combustion grate 2, primary air flow rate of the burnout grate, secondary air temperature, and secondary air flow rate; the output variable of the temperature proxy model is the average temperature of the outlet section of the third flue of the incinerator.

[0019] Compared with the prior art, the present invention has the following advantages and technical effects: The temperature surrogate model constructed in this invention can approximate the input-output mapping relationship of numerical simulation software with extremely high accuracy. The error between the model's prediction results and the numerical simulation results is within an acceptable range for engineering applications, fully ensuring the accuracy of temperature prediction. Simultaneously, this surrogate model belongs to a lightweight shallow neural network structure, with extremely low computational overhead for a single forward inference. The time to complete a temperature prediction is only in milliseconds, which is several orders of magnitude higher than the several hours of computation time of traditional numerical simulation software, meeting the stringent requirements of real-time data updates for incinerator digital twin systems. This invention uses interval type II fuzzy sets instead of traditional type I fuzzy sets, effectively characterizing and encompassing uncertainties introduced by factors such as waste composition fluctuations, boundary condition changes, and measurement noise. An adaptive shape factor is introduced into the membership function, enabling the model to adaptively adjust the shape of the membership function according to the distribution characteristics of the training samples, further enhancing the model's generalization performance and anti-interference ability under changing operating conditions. This invention requires numerical simulation calculations only during the training phase of the proxy model. Once the model training is complete, it can run in real time on a regular industrial computer without the need for a high-performance simulation server. This significantly reduces the hardware threshold and operating costs of digital twin technology. Furthermore, the model input variables are all controllable operating parameters that can be directly obtained from the industrial site through a distributed control system, eliminating the need for additional sensors and providing excellent engineering deployment convenience. This invention provides efficient and reliable technical support for the construction of digital twin systems for urban solid waste incineration plants, the optimization and control of the incineration process, and the intelligent operation and maintenance of equipment. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a comparison chart of the prediction results and numerical simulation results of the proxy model in this embodiment of the invention on the test set; Figure 3 This is an error graph showing the difference between the predicted and actual values ​​of the proxy model during the testing process of an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] Example 1 like Figure 1 As shown, this embodiment provides a method for predicting incinerator temperature based on numerical simulation and interval type II fuzzy neural network, including: Obtain the initial operating parameters of the incinerator, and establish a numerical simulation model of the incinerator based on the initial operating parameters; Orthogonal experimental design is performed on the controllable operating parameters in the numerical simulation model, and the numerical simulation model is run according to the orthogonal experimental design results to obtain simulation datasets under multiple working conditions. An interval type II fuzzy neural network is constructed, and the interval type II fuzzy neural network is trained using the simulation dataset to obtain a temperature proxy model; The target operating parameters are input into the temperature proxy model, and the temperature proxy model outputs the predicted temperature value at the incinerator outlet.

[0024] Furthermore, the initial operating parameters of the incinerator are obtained, and a numerical simulation model of the incinerator is established based on the initial operating parameters, including: Obtain actual operating data of the incinerator, including primary air temperature, primary air flow rate, grate speed, and combustion rate; Based on actual operating data, a numerical simulation model of the incinerator was established by coupling solid-phase combustion simulation software and gas-phase combustion simulation software. The actual operating data is used as the boundary conditions input into the numerical simulation model, and the simulation output results are compared and verified with the field measured values ​​to obtain a reliable numerical simulation model.

[0025] Specifically, the numerical simulation model adopts a dual-software coupled simulation approach using FLIC and Fluent. FLIC software models the combustion process of solid waste on the grate, with input variables including primary air temperature, primary air flow rate, grate velocity, incineration rate, and radiant heat distribution. Its outputs are solid phase temperature distribution, gas phase release rate, and radiant heat distribution. Fluent software models the combustion process of gaseous components within the furnace, with inputs being the gas phase release rate and radiant heat distribution output from FLIC. Its outputs are the furnace temperature field and flue gas component concentration field.

[0026] When solid-phase MSW burns on the grate, the interaction between the solid and gas phases is involved, and the conservation equation is shown below.

[0027] The solid continuity equation is: (1) in, f Porosity r s The density is the solid phase density, kg / m³. 3 , U s for x The velocity of the solid phase in the direction of the direction, m / s, V s for y The velocity of the solid phase in the direction of the direction, m / s, S m，dry , S m，devol and S m，char The conversion rates from solid to gas phase due to water evaporation, volatile matter release, and coke combustion are respectively expressed in kg / (m³). 3 ·s).

[0028] The solid-phase energy equation is: (2) in, H s The enthalpy of a solid, J / kg. l s The thermal dispersion coefficient of the solid is expressed in W / (m·K). T s For solid temperature, K. q px for x Radiative heat flux density in the direction, W / m q py for y Radiative heat flux density in the direction, W / m; S heat，exch and S heat，combThese are the heat exchange between the gas and solid phases, and the heat source for gas and coke combustion, respectively, in W / m³. 3 .

[0029] The gas phase continuity equation is (3) in, r g The density is in the gas phase, kg / m³ 3 , U g for x The velocity of the gas in the direction of travel, in m / s. V g for y The velocity of the gas in the direction of travel, in m / s.

[0030] The gas phase energy equation is: (4) in, H g The enthalpy of a gas, J / kg. l g The thermal dispersion coefficient of the gas is expressed in W / (mK). T g Let K be the gas temperature.

[0031] The transport equations for gas phase components are as follows: (5) in, D i，g gas i The diffusion coefficient, m 2 / s.

[0032] Partial reaction equations and kinetic parameters for the combustion of volatiles: (6) (7) in, C CO , and These are the mole fractions of CO, O2, and H2O, respectively.

[0033] (8) (9) in, The mole fraction of H2.

[0034] During the model construction process, a mesh model of the incinerator was first established using mesh generation tools such as ICEM, and the mesh independence was verified to ensure that the impact of changes in the number of meshes on key output variables was within the allowable error range in engineering. Then, using the operating data (including primary air temperature, primary air flow rate, grate velocity, incineration rate, and other boundary conditions) collected by the DCS of a MSWI plant in Beijing under actual stable operating conditions as input, the simulation output results were compared and verified with the field measured values ​​to confirm that the established numerical simulation model has engineering-acceptable credibility.

[0035] Furthermore, orthogonal experimental designs were performed on the controllable operating parameters in the numerical simulation model. Based on the results of the orthogonal experimental design, the numerical simulation model was run to obtain simulation datasets under multiple operating conditions, including: Multiple controllable operating parameters that affect the combustion effect are selected from the incinerator's operating conditions, and multiple level values ​​are determined for each controllable operating parameter. An orthogonal experimental design method is adopted to generate an orthogonal experimental scheme based on multiple controllable operating parameters and multiple level values ​​of each controllable operating parameter. The orthogonal experimental scheme includes multiple experimental conditions. The numerical simulation model was run sequentially according to each experimental condition in the orthogonal experimental scheme, and the temperature data at the incinerator outlet was recorded under each condition. The temperature data is preprocessed to obtain the simulation dataset.

[0036] Preprocessing includes outlier removal, normalization, and dividing the dataset into training, validation, and test sets.

[0037] The orthogonal experimental design is as follows: After the baseline operating condition model was validated, based on expert experience and simulation mechanism knowledge, 11 controllable factors that have the most significant impact on the incineration effect were selected from the many operable variables of the incinerator. These factors include: incineration rate (unit: t), grate velocity (unit: m / h), inlet air temperature of the drying grate (unit: K), inlet air temperature of the combustion section grate (unit: K), outlet air temperature of the primary air heater (unit: K), primary air flow rate of the drying grate (unit: Nm³ / h), primary air flow rate of combustion grate 1 (unit: Nm³ / h), primary air flow rate of combustion grate 2 (unit: Nm³ / h), primary air flow rate of the burnout grate (unit: Nm³ / h), secondary air temperature (unit: K), and secondary air flow rate (unit: Nm³ / h).

[0038] For the 11 factors mentioned above, four levels were determined for each factor based on its actual adjustable range. An orthogonal experimental design method was employed, using a suitable orthogonal array (or a larger-capacity mixed-level orthogonal array) to arrange the experimental scheme. The selection principle for the orthogonal array was: the number of columns in the orthogonal array should not be less than the number of factors, and the number of rows (i.e., the number of experiments) should be minimized while covering representative sample points for each factor and level combination. Based on the factor level combinations determined by the orthogonal array, simulation experiments were executed one by one in the numerical simulation module, recording key output variables such as the average temperature, O2 concentration, and CO2 concentration at the outlet section of the third flue of the incinerator under each operating condition, forming the original simulation dataset. Outlier removal and normalization preprocessing were performed on the collected data, and the dataset was divided into training, validation, and test sets according to a certain proportion.

[0039] Furthermore, the interval type II fuzzy neural network includes an input layer, a membership function layer, a rule layer, a consequent layer, and an output layer.

[0040] The functions of each layer in the interval type II fuzzy neural network are as follows: The input layer receives input variables and transmits the input variables to the membership function layer; The membership function layer uses the interval type II Gaussian membership function to fuzzify the input variables into interval-form membership values; The rule layer calculates the upper and lower emission intensities of each fuzzy rule based on the membership value; The backing layer performs type reduction based on the upper and lower emission intensities to obtain upper and lower output values; The output layer deblurs the upper and lower output values ​​and outputs the predicted temperature value at the incinerator outlet.

[0041] In this method, an adaptive shape factor is introduced into the interval type II Gaussian membership function. The adaptive shape factor is a positive parameter, which, together with the center and width of the interval type II Gaussian membership function, serves as the network parameter and is optimized and adjusted through a hierarchical learning algorithm.

[0042] The consequent layer uses the Begian-Melek-Mendel method for type reduction and deblurring.

[0043] The parameters of the interval type II fuzzy neural network are updated using a hierarchical learning algorithm, wherein the linear prior parameters and shape factor are learned using the least squares estimation method, and the nonlinear result parameters are learned using the gradient descent method.

[0044] Specifically, the temperature surrogate model (CI-IT2FNN surrogate model) is as follows: In IT2FNN, the IF-THEN rule is used to explain the relationship between the antecedent and the result. In the case of multiple inputs and a single output, each rule of IT2FNN can be represented as: rule : (10) after (11) in, , It is the number of fuzzy rules. It is the dimension of the input. For the nth interval type 2 fuzzy set corresponding to the kth input variable, It is a rule The output below.

[0045] The overall structure of IT2FNN includes an input layer, a membership function layer, a rule layer, a consequent layer, and an output layer. The functions of each layer are described below.

[0046] First layer: The number of nodes in this layer is the same as the number of input variables. The node's output is: (12) The second layer: The input variables are fuzzified and represented as intervals in this layer. This layer has... There are k nodes. Taking a Gaussian fuzzy membership function with uncertain width as an example, the output of the k-th node is: (13) (14) in, and Describes the membership functions of a binary fuzzy set. These represent the center width, upper width, and lower width of the membership function, respectively.

[0047] Layer 3: This layer calculates the emission intensity for each rule. This layer has... The node, the The output of each node is calculated as follows: (15) (16) in, and These represent the upper and lower emission intensities of the Kth node, respectively.

[0048] Level 4: This level performs type reduction. The output of this level is calculated as follows: (17) (18) in, and These represent the top and bottom outputs, respectively. and Determined by the switching point, and These represent the upper and lower weights of the Kth node, respectively.

[0049] Layer 5: This layer performs deblurring. Taking the center-average deblurr as an example, the output of this layer is: (19) This embodiment comprehensively improves IT2FNN by designing a fuzzy membership function with an adaptive factor to enhance the model's ability to handle uncertainty, thereby improving generalization performance. The efficiency of the defuzzification method and the hierarchical learning algorithm are effectively improved through the use of BMM. Finally, a hierarchical learning algorithm is employed to adjust the network parameters.

[0050] (1) The precise values ​​are fuzzified using the fuzzy membership function, and the unfuzziness of the fuzzy relationship is represented by the values ​​in the interval [0,1]. The Gaussian fuzzy membership function is widely used in IT2FNN, and its expression is: (20) in, For input values, and These represent the center and width of the function, respectively.

[0051] To improve the uncertainty representation capability of IT2FNN, the proposed method incorporates an adaptive shape factor into the Gaussian fuzzy membership function, as shown in the following equation: (twenty one) in, It is a positive number.

[0052] Like other parameters of the network, θ is learned through a hierarchical learning algorithm. Therefore, the shape of the fuzzy membership function can be optimized based on actual information.

[0053] (twenty two) (twenty three) (2) This method uses BMM for type reduction and defuzzification. At this point, the network has accurate subsequent weights, and the calculations for the fourth and fifth layers are given by the following formula: , (twenty four) (25) (26) (27) in, and Indicates the output of the emitter layer. and This represents the output of the result layer. This indicates the final output. This represents the weight of the Kth node in the result layer.

[0054] (3) Based on the hierarchical learning algorithm, the least squares estimation method is used to learn the linear prior parameters and shape factor. The nonlinear result parameters, i.e., the connection weights between the third and fourth layers, are learned through gradient descent. Therefore, while ensuring its generalization performance, the learning efficiency of IT2FNN is improved. The loss function for evaluating output accuracy is defined as: (28) in, This represents the network's output error in step five.

[0055] The update method for prior parameters and shape factors is expressed as follows: (29) The algorithm was trained using multi-condition data obtained from orthogonal experiments as the dataset.

[0056] like Figure 2 As shown, the prediction results of the constructed temperature surrogate model can effectively represent the results of the numerical simulation software. Figure 3 It can be seen that the fluctuation ranges of the error, absolute error, and relative error between the temperature surrogate model prediction results and the numerical simulation software simulation results are -23 to 14 K, 0 to 23 K, and 0 to 2.50%, respectively. In practical engineering applications, the above fluctuation values ​​are all within an acceptable range, indicating that the former can replace the latter and meet the engineering application needs of waste incineration power plants in real life, further assisting in the optimization control, operation and maintenance, and design and development of the waste incineration power generation process.

[0057] Compare CI-IT2FNN, Simplified Interval Type-2 Fuzzy Neural Network (SIT2FNN), RF, and FNN, and calculate their RMSE, MAE, and APE values ​​as evaluation metrics. Table 1 lists the parameter settings and comparison results.

[0058] Table 1 As shown in Table 1, the root mean square error (RMSE) of CI-IT2FNN on the training set is 9.9617, the mean absolute error (MAO) on the test set is 11.0426, the RMS error is 12.4987, and the mean percentage error (MPO) is 0.0121. CI-IT2FNN uses interval type-II fuzzy sets, which, compared with traditional single-type fuzzy sets, can better handle the uncertainty in the data, more flexibly describe and process the fuzziness and imprecision of the input data, and has strong learning efficiency and generalization ability.

[0059] Although CI-IT2FNN did not perform optimally across the various evaluation metrics in this experiment, the results may be influenced by a variety of factors, and it cannot be definitively concluded that it is optimal in all scenarios. For example, the characteristics of the experimental data, the model's parameter settings, and the randomness during training can all affect the results.

[0060] Compared with the prior art, the present invention has the following advantages: (i) High prediction accuracy. Experimental results show that the CI-IT2FNN temperature surrogate model constructed in this invention has a root mean square error (RMSE) of 12.4987 K, a mean absolute error (MAE) of 11.0426 K, and a mean absolute percentage error (MAPE) of 1.21% on the test set. The absolute error fluctuation range between the model's predicted values ​​and the numerical simulation software output values ​​is 0-23 K, and the relative error fluctuation range is 0-2.50%, both within the acceptable error range for engineering applications. This indicates that the surrogate model can accurately approximate the input-output mapping relationship of the numerical simulation software.

[0061] (II) Significantly Improved Computational Efficiency. The CI-IT2FNN surrogate model constructed in this invention belongs to a lightweight shallow neural network structure, with extremely low computational overhead for a single forward inference. In actual deployment, the model completes a temperature prediction in milliseconds, which is several orders of magnitude faster than the several hours required by traditional numerical simulation software. This meets the stringent requirements of the MSWI digital twin system for real-time data updates and provides a technically feasible path for online optimization control and real-time decision-making.

[0062] (III) Strong anti-interference ability and generalization performance. This invention uses interval type II fuzzy sets instead of traditional type I fuzzy sets. The membership function output is an interval rather than a precise value, which can effectively characterize and encompass the uncertainties introduced by factors such as fluctuations in waste composition, changes in boundary conditions, and measurement noise. An adaptive shape factor is introduced into the membership function, enabling the model to adaptively adjust the shape of the membership function according to the distribution characteristics of the training samples, further enhancing the model's generalization performance under changing operating conditions.

[0063] (iv) Strong engineering practicality. The technical paradigm of "numerical simulation generating data + online inference of surrogate model" proposed in this invention only requires numerical simulation calculations during the surrogate model training phase. Once the surrogate model is trained, it can run in real time on ordinary industrial computers or even embedded edge devices without the need for a high-performance simulation server, significantly reducing the hardware threshold and operating cost of digital twin technology. The model input consists of 11 directly obtainable and controllable factors from industrial sites (collected in real time through a DCS system), requiring no additional sensors and offering excellent engineering deployment convenience.

[0064] (v) Compact Model Size. This invention introduces a compact design strategy based on IT2FNN. By selecting a reasonable number of rules, it maintains a lightweight network structure while ensuring fitting accuracy. The model structure is simple and the number of parameters is controllable, avoiding the risk of overfitting and inference delay caused by an excessively large rule base, which is beneficial for running on resource-constrained industrial field equipment.

[0065] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting incinerator temperature based on numerical simulation and interval type-II fuzzy neural network, characterized in that, include: Obtain the initial operating parameters of the incinerator, and establish a numerical simulation model of the incinerator based on the initial operating parameters; Orthogonal experimental design is performed on the controllable operating parameters in the numerical simulation model, and the numerical simulation model is run according to the orthogonal experimental design results to obtain simulation datasets under multiple working conditions. An interval type II fuzzy neural network is constructed, and the interval type II fuzzy neural network is trained using the simulation dataset to obtain a temperature proxy model; The target operating parameters are input into the temperature proxy model, and the temperature proxy model outputs the predicted temperature value at the incinerator outlet.

2. The method according to claim 1, characterized in that, The process of obtaining the initial operating parameters of the incinerator and establishing a numerical simulation model of the incinerator based on the initial operating parameters includes: Obtain actual operating data of the incinerator, including primary air temperature, primary air flow rate, grate speed, and combustion rate; Based on the actual operating data, a numerical simulation model of the incinerator was established by coupling solid-phase combustion simulation software and gas-phase combustion simulation software. The actual operating data is used as boundary conditions to input the numerical simulation model, and the simulation output results are compared and verified with the field measured values ​​to obtain a reliable numerical simulation model.

3. The method according to claim 1, characterized in that, Orthogonal experimental design is performed on the controllable operating parameters in the numerical simulation model. The numerical simulation model is then run based on the orthogonal experimental design results to obtain simulation datasets under multiple operating conditions, including: Multiple controllable operating parameters that affect the combustion effect are selected from the incinerator's operating conditions, and multiple level values ​​are determined for each controllable operating parameter. An orthogonal experimental design method is adopted to generate an orthogonal experimental scheme based on the multiple controllable operating parameters and multiple level values ​​of each controllable operating parameter. The orthogonal experimental scheme includes multiple experimental conditions. The numerical simulation model was run sequentially according to each experimental condition in the orthogonal experimental scheme, and the temperature data at the incinerator outlet was recorded under each condition. The temperature data is preprocessed to obtain a simulation dataset.

4. The method according to claim 3, characterized in that, The preprocessing includes outlier removal, normalization, and dividing the dataset into training, validation, and test sets.

5. The method according to claim 1, characterized in that, The interval type II fuzzy neural network includes an input layer, a membership function layer, a rule layer, a consequent layer, and an output layer.

6. The method according to claim 5, characterized in that, The functions of each layer in the interval type-II fuzzy neural network are as follows: The input layer receives input variables and transmits the input variables to the membership function layer; The membership function layer uses the interval type II Gaussian membership function to fuzzify the input variables into interval-form membership values; The rule layer calculates the upper and lower emission intensities of each fuzzy rule based on the membership value; The backing layer performs type reduction based on the upper and lower emission intensities to obtain upper and lower output values; The output layer deblurs the upper and lower output values ​​and outputs the predicted temperature value at the incinerator outlet.

7. The method according to claim 6, characterized in that, The interval type II Gaussian membership function introduces an adaptive shape factor, which is a positive parameter. This adaptive shape factor, together with the center and width of the interval type II Gaussian membership function, serves as a network parameter and is optimized and adjusted through a hierarchical learning algorithm.

8. The method according to claim 6, characterized in that, The consequent layer is type-reduction and deblurring performed using the Begian-Melek-Mendel method.

9. The method according to claim 7, characterized in that, The parameters of the interval type II fuzzy neural network are updated using a hierarchical learning algorithm, wherein the linear prior parameters and shape factor are learned using the least squares estimation method, and the nonlinear result parameters are learned using the gradient descent method.

10. The method according to claim 1, characterized in that, The input variables of the temperature proxy model are the controllable operating parameters of the operating conditions, including combustion rate, grate speed, inlet air temperature of the drying grate, inlet air temperature of the combustion grate, outlet air temperature of the primary air heater, primary air flow rate of the drying grate, primary air flow rate of combustion grate 1, primary air flow rate of combustion grate 2, primary air flow rate of the burnout grate, secondary air temperature, and secondary air flow rate; the output variable of the temperature proxy model is the average temperature of the outlet section of the third flue of the incinerator.

Images