Intelligent prediction method and system for temperature in incinerator based on transfer learning

Abstract

The invention relates to the technical field of solid treatment, in particular to an intelligent prediction method and system for the temperature in an incinerator based on transfer learning. A transfer learning thought is utilized, existing incineration condition data are fully utilized to train a general BP neural network model, rapid adaptation can be achieved only through small-amount fine tuning for new conditions, and the prediction precision under the condition of insufficient samples is greatly improved. According to the method, operation parameters and in-furnace temperature data under different working conditions are introduced, a simulation data pre-training set and a field data target set are constructed, a BP neural network and a transfer learning method are combined, and a temperature distribution rule in historical data is fully utilized. A complex model does not need to be trained from the beginning, fine adjustment is carried out on the basis of an existing model, the modeling efficiency is improved, the model interpretability is enhanced, and analysis and diagnosis in engineering application are facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of solid disposal, and in particular to an intelligent prediction method and system for the temperature in an incinerator based on transfer learning. Background Art

[0002] With the rapid development of urbanization, the output of urban domestic waste has been rising year by year. According to the latest statistics, the total output of urban domestic waste in the world is expected to reach 2.7 billion tons in 2024, and China will produce 1.7 billion tons. Urban domestic waste refers to solid waste generated in the daily life activities of urban residents. Its main components include food residues, paper, plastics, fabrics, etc., with high organic content and complex and diverse components. During the stacking process, domestic waste is prone to rot and ferment, producing odorous gases, leachate, etc., which may cause serious pollution of soil, groundwater and atmosphere, and have an adverse impact on the surrounding environmental quality and residents' lives. Domestic waste incineration is an important means of urban solid waste treatment, which can effectively reduce the amount of waste and recycle the heat energy generated during the incineration process. Accurate prediction of the internal temperature of the domestic waste incinerator is crucial to controlling the incineration process, improving energy efficiency and ensuring incineration safety.

[0003] Traditional methods for predicting the temperature inside a municipal solid waste incinerator are usually based on physical models, which require a large amount of process parameter and environmental condition data as input. The model is highly complex and the prediction accuracy needs to be further improved. In recent years, machine learning technology has shown good performance in industrial process modeling and prediction. Among them, BP neural network has become an effective choice for predicting the temperature inside a municipal solid waste incinerator with its powerful nonlinear modeling ability and good generalization ability. However, a single BP neural network model may have performance degradation when dealing with temperature prediction tasks under different working conditions, and a large number of samples are required. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that a single BP neural network model may have a performance degradation problem when processing temperature prediction tasks under different working conditions, and a large number of samples are required.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides an intelligent prediction method for the temperature in an incinerator based on transfer learning, which includes obtaining numerically simulated incinerator temperature data, establishing a pre-training data set, collecting actual incinerator operation data, and establishing a target data set;

[0008] Build a BP neural network model based on the pre-training data set, train the model using the pre-training data set, and save the weight parameters of the pre-training model;

[0009] Freeze all layers of the pre-trained model except the last hidden layer, add a new hidden layer after the frozen layer, and fine-tune the newly added layer using the target dataset;

[0010] Changing the number of hidden layers and hyperparameter configurations, repeating the transfer learning process, and selecting the model with the best prediction effect, wherein the model with the best prediction effect has the smallest loss function;

[0011] Use the optimized transfer learning model to predict the temperature at different locations in the incinerator under different operating conditions.

[0012] As a preferred solution of the intelligent prediction method of the temperature in the incinerator based on transfer learning of the present invention, wherein: the acquisition of numerically simulated incinerator temperature data and the establishment of a pre-training data set include:

[0013] Obtain the temperature data of domestic waste and different positions in the furnace for numerical solution under different working conditions, and establish a data set for BP neural network pre-training;

[0014] Collect temperature data of domestic waste and different positions in the incinerator under different working conditions during actual operation of the incinerator manufacturer, and establish a target dataset for transfer learning fine-tuning;

[0015] Perform simulation pre-training on the data set to obtain the initial weights of the BP neural network and save them for subsequent transfer learning;

[0016] Freeze the neural network topology and weights of all layers except the last layer in the pre-trained model to obtain the initial BP neural network weights for transfer learning;

[0017] Based on the target data set, migrate the topological structure and weight data saved by the frozen pre-trained model to fine-tune the topological structure of the BP neural network;

[0018] Repeat the transfer learning process to find a set of neural networks that has the best prediction effect on the target data set, that is, the smallest loss function;

[0019] The optimal prediction model of BP neural network based on transfer learning is used to predict the temperature at different positions in the incinerator under different working conditions.

[0020] As a preferred solution of the method for predicting the temperature at different positions in the incinerator under different working conditions described in the present invention, wherein: the acquisition of the numerically solved domestic waste and the temperature data at different positions in the furnace under different working conditions includes pre-processing the data after pyrolysis and gasification of domestic waste in the incinerator under different working conditions as a data set for pre-training of the BP neural network, specifically including the following steps:

[0021] Select factors that affect temperature change such as moisture content, volatile matter, fixed carbon, ash, garbage particle size, primary air, circulating air, secondary air, load, etc. of different domestic garbage under different operating parameters as input variables, and normalize all factors;

[0022] Based on the average influence value algorithm, all influencing factors are reduced in dimension;

[0023] The input of the BP neural network is expressed as:

[0024] X(i)=[x(1),x(2),...,x(i),x(n)],i∈[1,n]

[0025] x(i)=[x1(i),x2(i),...,xm(i)],i∈[1,n]

[0026] Among them, X(i) represents the sequence of influencing factors under different working conditions i; x(i) is a vector composed of m different factors; the temperatures at different locations in the furnace used for prediction under different working conditions are as follows:

[0027] Y(i)=[y(1),y(2),...,y(i),...,y(n)],i∈[1,n]

[0028] Among them, Y(i) represents the temperature sequence of different positions in the furnace under different working conditions i, that is, for n different working conditions, each working condition i has an output vector y(i) composed of the temperatures at different positions in the furnace;

[0029] For different working conditions i, the characteristic variables in x(i) are normalized, and the input features x(i) under each working condition are mapped to the interval [0,1] to eliminate dimensional differences and improve the stability and convergence of model training;

[0030] The dataset used for training is constructed as follows:

[0031] A={X(i),Y(i)}={(x(1),y(1)),(x(2),y(2)),(x(i),y(i)),…,(x(n),y(n))}

[0032] Among them, (x(i), y(i)) represents the normalized data pair under the i-th working condition;

[0033] Based on different working conditions, dataset A is further divided into pre-training dataset and prediction dataset.

[0034] As a preferred solution of the intelligent prediction method of the temperature in the incinerator based on transfer learning of the present invention, the on-site collection of temperature data of domestic garbage and different positions in the incinerator under different working conditions in the actual operation of a certain incinerator manufacturer includes repeating the following steps:

[0035] The factors that affect the temperature change under different working conditions, such as the moisture content, volatile matter, fixed carbon, ash, garbage particle size, primary air, circulating air, secondary air, and load of different domestic garbage are selected as input variables, and all factors are normalized;

[0036] Based on the average influence value algorithm, all influencing factors are reduced in dimension;

[0037] The input of the BP neural network is expressed as:

[0038] X(i)=[x(1),x(2),...,x(i),x(n)],i∈[1,n]

[0039] x(i)=[x1(i),x2(i),...,xm(i)],i∈[1,n]

[0040] Among them, X(i) represents the sequence of influencing factors under different working conditions i; x(i) is a vector composed of m different factors;

[0041] The temperatures at different locations in the furnace used for prediction under different working conditions are as follows:

[0042] Y(i)=[y(1),y(2),...,y(i),...,y(n)],i∈[1,n]

[0043] Among them, Y(i) represents the temperature sequence of different positions in the furnace under different working conditions i, that is, for n different working conditions, each working condition i has an output vector y(i) composed of the temperatures at different positions in the furnace;

[0044] For different working conditions i, the characteristic variables in x(i) are normalized, and the input features x(i) under each working condition are mapped to the interval [0,1] to eliminate dimensional differences and improve the stability and convergence of model training;

[0045] The dataset used for training is constructed as follows:

[0046] A={X(i),Y(i)}={(x(1),y(1)),(x(2),y(2)),(x(i),y(i)),…,(x(n),y(n))}

[0047] Among them, (x(i), y(i)) represents the normalized data pair under the i-th working condition;

[0048] Based on different working conditions, dataset A is further divided into pre-training dataset and prediction dataset.

[0049] As a preferred solution of the method for predicting the temperature at different positions in the incinerator under different working conditions described in the present invention, wherein: the data set is simulated and pre-trained to obtain the initialization weights of the BP neural network, which are saved for subsequent transfer learning, and the initialization training adopts the following steps:

[0050] According to the BP neural network architecture, the hyperparameter configuration HyperParameter is optimized, and j hyperparameter configuration experiments are set. Each experiment corresponds to a different set of hyperparameter configurations, recorded as HyperParameter k ; where k represents the kth experiment, and its value range is k∈[1,j];

[0051] The pre-trained data set is set to a ratio of 8:2, randomly divided into training set and test set, and brought into the BP neural network for forward propagation calculation respectively;

[0052] The BP neural network adopts the encoder-decoder architecture design. The encoder and decoder are respectively composed of several stacked transformer blocks. During pre-training, the training set X(i) is first positionally encoded through the encoder part, and then input into the encoder block of the transformer neural network for feature encoding, and finally the output data of the encoder is obtained;

[0053] During pre-training, Y(i) is position-encoded and then input into the decoder. The first transformer decoder block of the decoder network masks the input position-encoded target output data Y(i) and substitutes it and the encoder output data into the subsequent decoder block for training. The masked decoder input is interactively fused with the feature representation output by the encoder network, and the cross-attention mechanism is used to generate the target output sequence for training.

[0054] Training is completed, save the final pre-trained BP neural network weight parameter configuration Pt k , where k represents the kth weight parameter configuration;

[0055] The neural network topology and weights of all layers except the last layer in the pre-trained model are frozen to obtain the initial BP neural network weights for transfer learning, so that its weight parameters do not change during training, and the BP neural network after transfer learning is obtained.

[0056] As a preferred solution of the method for predicting the temperature at different positions in the incinerator under different working conditions described in the present invention, the topological structure and weight data saved by the migration and freezing pre-training model are fine-tuned based on the target data set, and the specific steps of fine-tuning are as follows:

[0057] The target data set collected on site was set to a ratio of 8:2 and randomly divided into a training set and a test set;

[0058] The target training set and test set are input into the training according to the initialization training method. The fine-tuning method only trains and modifies the hyperparameter configuration in the transfer learning architecture. At the same time, the BP neural network weight parameter configuration after fine-tuning transfer learning is saved, which is recorded as Pt k ′;

[0059] The four evaluation indicators of root mean square error, mean absolute percentage error, absolute percentage error and determination coefficient are selected to evaluate the performance of the BP neural network model after transfer learning, which are defined as follows:

[0060]

[0061] Among them, RMSE represents the root mean square error, MAPE represents the mean absolute percentage error, APE represents the absolute percentage error and R 2 represents the coefficient of determination; n represents a total of n prediction samples, y(i) and They represent the actual value of the i-th sample and the predicted value of the model under n different working conditions respectively; RMSE, MAPE, APE and R2 are sorted and saved in the error set Ci, which can be expressed as:

[0062] C i =(RMSE i 、MAPE i 、APE、R 2 )

[0063] Among them, RMSE represents the root mean square error, MAPE represents the mean absolute percentage error, APE represents the absolute percentage error and R 2 represents the coefficient of determination.

[0064] As a preferred solution of the method for predicting the temperature at different positions in the incinerator under different working conditions described in the present invention, the specific steps of repeating the transfer learning process to find a group of neural networks with the best prediction effect on the target data set, i.e., the smallest loss function, are as follows:

[0065] Change the hyperparameter configuration, recorded as the hyperparameter configuration HyperParameter i , where i∈[2,j];

[0066] Repeated BP neural network adopts the encoder-decoder architecture design-fine-tuning specific steps to obtain the weight parameter configuration Pt in turn k , the fine-tuned weight parameter configuration Pt k ′, error set D k ;

[0067] Repeat the transfer learning process twice until the number of hyperparameter configuration debugging experiments reaches j times, and find the group with the smallest error in j experiments, which is recorded as D best , the corresponding fine-tuning parameter weight is recorded as Pt best ′, the hyperparameter configuration is recorded as HyperParameter best , that is, the hyperparameter configuration with the best model prediction effect is obtained;

[0068] The effect of the BP neural network without transfer learning was compared with that of the BP neural network after transfer learning. The predicted values ​​and error sets obtained by the two algorithms were compared and analyzed. Finally, the optimal prediction model of transfer learning based on the BP neural network was obtained as a prediction method for the temperature at different positions in the domestic waste incinerator under different working conditions.

[0069] In a second aspect, an embodiment of the present invention provides an intelligent prediction system for the temperature in an incinerator based on transfer learning, which includes an acquisition module for acquiring numerically simulated incinerator temperature data, establishing a pre-training data set, collecting actual incinerator operation data, and establishing a target data set;

[0070] The pre-training module builds a BP neural network model based on the pre-training data set, trains the model using the pre-training data set, and saves the weight parameters of the pre-training model;

[0071] Freeze module, freeze all layers of the pre-trained model except the last hidden layer, add a new hidden layer after the frozen layer, and use the target dataset to fine-tune the newly added layer;

[0072] Iterate the module, change the number of hidden layers and hyperparameter configuration, repeat the transfer learning process, and select the model with the best prediction effect, wherein the model with the best prediction effect has the smallest loss function;

[0073] The prediction module uses the optimized transfer learning model to predict the temperature at different locations in the incinerator under different operating conditions.

[0074] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the intelligent prediction method for the temperature in an incinerator based on transfer learning as described in the first aspect of the present invention are implemented.

[0075] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the intelligent prediction method for the temperature in an incinerator based on transfer learning as described in the first aspect of the present invention are implemented.

[0076] The beneficial effects of the present invention are as follows: This scheme predicts the temperature at different positions in a domestic waste incinerator under different working conditions through transfer learning based on BP neural network. This method uses the idea of ​​transfer learning and fully utilizes the existing incineration working condition data to train a general BP neural network model. For new working conditions, only limited fine-tuning training is required to quickly adapt to the temperature prediction needs under different working conditions. The core of this method is to extract basic knowledge from a large-scale / low-cost source domain and apply it to a small-scale / high-cost target domain. This greatly improves the problem of insufficient task prediction accuracy caused by insufficient samples.

[0077] 2. The present invention introduces the operating parameters of the domestic waste incinerator under different working conditions and different temperatures in the furnace to construct a simulation data pre-training set and a target set of field collected data for fine-tuning, introduces the BP neural network architecture, and uses its excellent nonlinear fitting ability to train to obtain hyperparameter configuration; by changing the hyperparameter configuration, a set of hyperparameter configurations with the best training effect is found; in combination with the transfer learning method, the target data set is substituted into the BP neural network after transfer learning for fine-tuning, which can make full use of the temperature distribution law contained in the historical data; based on the fine-tuned transfer learning BP neural network, the temperature at different positions in the incinerator is predicted; a control group is used to predict the temperature at different positions in the furnace under different working conditions, and the model prediction effect is evaluated based on the four indicators of RMSE, MAPE, APE and R2, which proves from a quantitative perspective that the model based on transfer learning and BP neural network architecture is the optimal model in this task, which can be used for temperature prediction at different positions in the furnace under different working conditions, and effectively improves the prediction accuracy of the temperature at different positions in the incinerator.

[0078] 3. The present invention adopts a transfer learning model based on the BP neural network architecture, which eliminates the need to train a complex BP neural network model from scratch; limited fine-tuning is performed on the basis of the existing model, which greatly improves the efficiency of modeling; the interpretability of the model can be increased to a certain extent by visually analyzing the feature extraction process of the hidden layer; and it can more easily cope with unknown temperature fluctuations in the incinerator, which is beneficial to analysis and diagnosis in engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0080] Figure 1 It is a flow chart of the method for predicting the temperature at different positions in the incinerator under different working conditions;

[0081] Figure 2 A computer equipment diagram for predicting the temperature at different locations in the incinerator under different operating conditions;

[0082] Figure 3 A schematic diagram of a three-dimensional calculation model of a domestic waste incinerator for predicting the temperature at different locations in the incinerator under different operating conditions;

[0083] Figure 4 The error graph for training the BP neural network framework with data input for the prediction method of the temperature at different locations in the incinerator under different operating conditions;

[0084] Figure 5 The error diagram of the BP neural network architecture trained after data input transfer learning for the prediction method of the temperature at different locations in the incinerator under different operating conditions. DETAILED DESCRIPTION

[0085] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0086] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0087] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0088] Example 1

[0089] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides an intelligent prediction method for the temperature in an incinerator based on transfer learning, comprising:

[0090] S100: Acquire the temperature data of the incinerator for numerical simulation, establish a pre-training data set, collect the actual incinerator operation data, and establish a target data set;

[0091] S101: Obtain the temperature data of the incinerator for numerical simulation and establish a pre-training data set including:

[0092] Obtain the temperature data of domestic waste and different positions in the furnace for numerical solution under different working conditions, and establish a data set for BP neural network pre-training;

[0093] Collect temperature data of domestic waste and different positions in the incinerator under different working conditions during actual operation of the incinerator manufacturer, and establish a target dataset for transfer learning fine-tuning;

[0094] Perform simulation pre-training on the data set to obtain the initial weights of the BP neural network and save them for subsequent transfer learning;

[0095] Freeze the neural network topology and weights of all layers except the last layer in the pre-trained model to obtain the initial BP neural network weights for transfer learning;

[0096] Based on the target data set, migrate the topological structure and weight data saved by the frozen pre-trained model to fine-tune the topological structure of the BP neural network;

[0097] Repeat the transfer learning process to find a set of neural networks that has the best prediction effect on the target data set, that is, the smallest loss function;

[0098] The optimal prediction model of BP neural network based on transfer learning is used to predict the temperature at different positions in the incinerator under different working conditions.

[0099] S102: obtaining the numerically solved data of the domestic waste and the temperature data at different positions in the furnace under different working conditions, including pre-processing the data of the domestic waste after pyrolysis and gasification in the incinerator under different working conditions, as a data set for pre-training of the BP neural network, specifically including the following steps:

[0100] Select factors that affect temperature change such as moisture content, volatile matter, fixed carbon, ash, garbage particle size, primary air, circulating air, secondary air, load, etc. of different domestic garbage under different operating parameters as input variables, and normalize all factors;

[0101] Based on the average influence value algorithm, all influencing factors are reduced in dimension;

[0102] The input of the BP neural network is expressed as:

[0103] X(i)=[x(1),x(2),...,x(i),x(n)],i∈[1,n]

[0104] x(i)=[x1(i),x2(i),...,xm(i)],i∈[1,n]

[0105] Among them, X(i) represents the sequence of influencing factors under different working conditions i; x(i) is a vector composed of m different factors; the temperatures at different locations in the furnace used for prediction under different working conditions are as follows:

[0106] Y(i)=[y(1),y(2),...,y(i),...,y(n)],i∈[1,n]

[0107] Among them, Y(i) represents the temperature sequence of different positions in the furnace under different working conditions i, that is, for n different working conditions, each working condition i has an output vector y(i) composed of the temperatures at different positions in the furnace;

[0108] For different working conditions i, the characteristic variables in x(i) are normalized, and the input features x(i) under each working condition are mapped to the interval [0,1] to eliminate dimensional differences and improve the stability and convergence of model training;

[0109] The dataset used for training is constructed as follows:

[0110] A={X(i),Y(i)}={(x(1),y(1)),(x(2),y(2)),(x(i),y(i)),…,(x(n),y(n))}

[0111] Among them, (x(i), y(i)) represents the normalized data pair under the i-th working condition;

[0112] Based on different working conditions, dataset A is further divided into pre-training dataset and prediction dataset.

[0113] S103: On-site collection of temperature data of domestic waste and different positions in the incinerator under different operating conditions during the actual operation of an incinerator manufacturer, including repeating the following steps:

[0114] The factors that affect the temperature change under different working conditions, such as the moisture content, volatile matter, fixed carbon, ash, garbage particle size, primary air, circulating air, secondary air, and load of different domestic garbage are selected as input variables, and all factors are normalized;

[0115] Based on the average influence value algorithm, all influencing factors are reduced in dimension;

[0116] The input of the BP neural network is expressed as:

[0117] X(i)=[x(1),x(2),...,x(i),x(n)],i∈[1,n]

[0118] x(i)=[x1(i),x2(i),...,xm(i)],i∈[1,n]

[0119] Among them, X(i) represents the sequence of influencing factors under different working conditions i; x(i) is a vector composed of m different factors;

[0120] The temperatures at different locations in the furnace used for prediction under different working conditions are as follows:

[0121] Y(i)=[y(1),y(2),...,y(i),...,y(n)],i∈[1,n]

[0122] Among them, Y(i) represents the temperature sequence of different positions in the furnace under different working conditions i, that is, for n different working conditions, each working condition i has an output vector y(i) composed of the temperatures at different positions in the furnace;

[0123] For different working conditions i, the characteristic variables in x(i) are normalized, and the input features x(i) under each working condition are mapped to the interval [0,1] to eliminate dimensional differences and improve the stability and convergence of model training;

[0124] The dataset used for training is constructed as follows:

[0125] A={X(i),Y(i)}={(x(1),y(1)),(x(2),y(2)),(x(i),y(i)),…,(x(n),y(n))}

[0126] Among them, (x(i), y(i)) represents the normalized data pair under the i-th working condition;

[0127] Based on different working conditions, dataset A is further divided into pre-training dataset and prediction dataset.

[0128] S200: constructing a BP neural network model based on a pre-training data set, training the model using the pre-training data set, and saving weight parameters of the pre-training model;

[0129] S201: Perform simulation pre-training on the data set to obtain the initialization weights of the BP neural network, which are saved for subsequent transfer learning. The initialization training adopts the following steps:

[0130] According to the BP neural network architecture, the hyperparameter configuration HyperParameter is optimized, and j hyperparameter configuration experiments are set. Each experiment corresponds to a different set of hyperparameter configurations, recorded as HyperParameter k ; where k represents the kth experiment, and its value range is k∈[1,j];

[0131] The pre-trained data set is set to a ratio of 8:2, randomly divided into training set and test set, and brought into the BP neural network for forward propagation calculation respectively;

[0132] The BP neural network adopts the encoder-decoder architecture design. The encoder and decoder are respectively composed of several stacked transformer blocks. During pre-training, the training set X(i) is first positionally encoded through the encoder part, and then input into the encoder block of the transformer neural network for feature encoding, and finally the output data of the encoder is obtained;

[0133] During pre-training, Y(i) is position-encoded and then input into the decoder. The first transformer decoder block of the decoder network masks the input position-encoded target output data Y(i) and substitutes it and the encoder output data into the subsequent decoder block for training. The masked decoder input is interactively fused with the feature representation output by the encoder network, and the cross-attention mechanism is used to generate the target output sequence for training.

[0134] Training is completed, save the final pre-trained BP neural network weight parameter configuration Pt k , where k represents the kth weight parameter configuration;

[0135] Freeze the neural network topology and weights of all layers except the last layer in the pre-trained model, obtain the initial BP neural network weights for transfer learning, so that its weight parameters do not change during training, and obtain the BP neural network after transfer learning.

[0136] S300: Freeze all layers of the pre-trained model except the last hidden layer, add a new hidden layer after the frozen layer, and fine-tune the newly added layer using the target dataset;

[0137] S301: Based on the target data set, migrate the topological structure and weight data saved by the frozen pre-training model, and fine-tune the topological structure of the BP neural network. The specific steps of fine-tuning are as follows:

[0138] The target data set collected on site was set to a ratio of 8:2 and randomly divided into a training set and a test set;

[0139] The target training set and test set are input into the training according to the initialization training method. The fine-tuning method only trains and modifies the hyperparameter configuration in the transfer learning architecture. At the same time, the BP neural network weight parameter configuration after fine-tuning transfer learning is saved, which is recorded as Pt k ′;

[0140] The four evaluation indicators of root mean square error, mean absolute percentage error, absolute percentage error and determination coefficient are selected to evaluate the performance of the BP neural network model after transfer learning, which are defined as follows:

[0141]

[0142] Among them, RMSE represents the root mean square error, MAPE represents the mean absolute percentage error, APE represents the absolute percentage error and R 2 represents the coefficient of determination; n represents a total of n prediction samples, y(i) and They represent the actual value of the i-th sample and the predicted value of the model under n different working conditions respectively; RMSE, MAPE, APE and R2 are sorted and saved in the error set Ci, which can be expressed as:

[0143] C i =(RMSE i 、MAPE i 、APE、R 2 )

[0144] Among them, RMSE represents the root mean square error, MAPE represents the mean absolute percentage error, APE represents the absolute percentage error and R 2 represents the coefficient of determination.

[0145] S400: Change the number of hidden layers and hyperparameter configuration, repeat the transfer learning process, select the model with the best prediction effect, and select the model with the best prediction effect with the smallest loss function;

[0146] S401: Repeat the transfer learning process to find a set of neural networks with the best prediction effect on the target data set, that is, the smallest loss function. The specific steps are as follows:

[0147] Change the hyperparameter configuration, recorded as the hyperparameter configuration HyperParameter i , where i∈[2,j];

[0148] Repeated BP neural network adopts the encoder-decoder architecture design-fine-tuning specific steps to obtain the weight parameter configuration Pt in turn k , the fine-tuned weight parameter configuration Pt k ′, error set D k ;

[0149] Repeat the transfer learning process twice until the number of hyperparameter configuration debugging experiments reaches j times, and find the group with the smallest error in j experiments, which is recorded as D best , the corresponding fine-tuning parameter weight is recorded as Pt best ′, the hyperparameter configuration is recorded as HyperParameter best , that is, the hyperparameter configuration with the best model prediction effect is obtained.

[0150] S500: Use the optimized transfer learning model to predict the temperature at different locations in the incinerator under different operating conditions.

[0151] S501: Compare the effect of the BP neural network without transfer learning with that of the BP neural network after transfer learning, compare and analyze the predicted values ​​and error sets obtained by the two algorithms, and finally obtain the optimal prediction model of transfer learning based on the BP neural network as a prediction method for the temperature at different positions in the domestic waste incinerator under different working conditions.

[0152] Furthermore, this embodiment also provides an intelligent prediction system for the temperature in an incinerator based on transfer learning, comprising:

[0153] The acquisition module acquires the temperature data of the incinerator for numerical simulation, establishes a pre-training data set, collects the actual incinerator operation data, and establishes a target data set;

[0154] The pre-training module builds a BP neural network model based on the pre-training data set, trains the model using the pre-training data set, and saves the weight parameters of the pre-training model;

[0155] Freeze module, freeze all layers of the pre-trained model except the last hidden layer, add a new hidden layer after the frozen layer, and use the target dataset to fine-tune the newly added layer;

[0156] Iterate the module, change the number of hidden layers and hyperparameter configuration, repeat the transfer learning process, select the model with the best prediction effect, and the model with the best prediction effect has the smallest loss function;

[0157] The prediction module uses the optimized transfer learning model to predict the temperature at different locations in the incinerator under different operating conditions.

[0158] This embodiment also provides a computer device, which is suitable for the case of an intelligent prediction method for the temperature inside an incinerator based on transfer learning, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent prediction method for the temperature inside an incinerator based on transfer learning as proposed in the above embodiment.

[0159] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0160] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for intelligently predicting the temperature in an incinerator based on transfer learning as proposed in the above embodiment is implemented.

[0161] In summary, this scheme predicts the temperature at different locations in the domestic waste incinerator under different working conditions through transfer learning based on BP neural network. This method uses the idea of ​​transfer learning and fully utilizes the existing incineration working condition data to train a general BP neural network model. For new working conditions, only limited fine-tuning training is required to quickly adapt to the temperature prediction needs under different working conditions. The core of this method is to extract basic knowledge from large-scale / low-cost source domains and apply it to small-scale / high-cost target domains. This greatly improves the problem of insufficient task prediction accuracy caused by insufficient samples.

[0162] 2. The present invention introduces the operating parameters of the domestic waste incinerator under different working conditions and different temperatures in the furnace to construct a simulation data pre-training set and a target set of field collected data for fine-tuning, introduces the BP neural network architecture, and uses its excellent nonlinear fitting ability to train to obtain hyperparameter configuration; by changing the hyperparameter configuration, a set of hyperparameter configurations with the best training effect is found; in combination with the transfer learning method, the target data set is substituted into the BP neural network after transfer learning for fine-tuning, which can make full use of the temperature distribution law contained in the historical data; based on the fine-tuned transfer learning BP neural network, the temperature at different positions in the incinerator is predicted; a control group is used to predict the temperature at different positions in the furnace under different working conditions, and the model prediction effect is evaluated based on the four indicators of RMSE, MAPE, APE and R2, which proves from a quantitative perspective that the model based on transfer learning and BP neural network architecture is the optimal model in this task, which can be used for temperature prediction at different positions in the furnace under different working conditions, and effectively improves the prediction accuracy of the temperature at different positions in the incinerator.

[0163] 3. The present invention adopts a transfer learning model based on the BP neural network architecture, which eliminates the need to train a complex BP neural network model from scratch; limited fine-tuning is performed on the basis of the existing model, which greatly improves the efficiency of modeling; the interpretability of the model can be increased to a certain extent by visually analyzing the feature extraction process of the hidden layer; and it can more easily cope with unknown temperature fluctuations in the incinerator, which is beneficial to analysis and diagnosis in engineering applications.

[0164] Example 2

[0165] Reference Figure 2 - Figure 5 , which is the second embodiment of the present invention, and this embodiment provides a method for predicting the temperature at different positions in an incinerator under different working conditions. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0166] Step S100: Figure 3 As shown, a proportional three-dimensional calculation model of a domestic waste incinerator is established, and numerical software is used to simulate it, and the simulation calculation data of the pyrolysis and gasification of domestic waste in the incinerator is obtained. The data is pre-processed and used as a data set for pre-training of the BP neural network;

[0167] According to the temperature data of different working conditions and different locations, the specific definition of the pre-training data set is as follows:

[0168] A={(X(i), Y(i))}={(x(1), y(1)), (x(2), y(2)), (x(i), y(i)), ..., (x(n), y(n))}i∈[1,n]

[0169] X(i)=[x(1),x(2),…,x(i),…,x(n)]i∈[1,n]

[0170] x(i)=[x1(i),x2(i),...,x m (i)],i∈[1,n]

[0171] Y(i)=[y(1),y(2),…,y(i),…,y(n)]i∈[1,n]

[0172] Among them, X(i) represents the sequence of influencing factors under different working conditions; x(i) is a vector composed of m different factors. Y(i) represents the temperature sequence of different positions in the furnace under different working conditions, that is, for n different working conditions, each condition has an output vector y(i) composed of the temperatures at different positions in the furnace;

[0173] Step S20000: This experiment cooperated with a domestic waste incineration plant in Shanghai to obtain an incinerator of the same model as the calculation model, and obtained and recorded the on-site data collected during the pyrolysis and gasification of domestic waste in the incinerator. The data was preprocessed in the same way as step S100 and used as the target data set for transfer learning fine-tuning.

[0174] Step S300: Pre-training is performed based on the simulation data set in S100 to obtain the initialization weights of the BP neural network, and the weights are saved for subsequent transfer learning;

[0175] like Figure 1 As shown in step S300 and Table 1, the specific process of BP neural network pre-training in step S300 is as follows:

[0176] Table 1 BP neural network pseudo code

[0177]

[0178] Step S301: Optimize the hyperparameter configuration HyperParameter according to the BP neural network architecture, set j hyperparameter configuration experiments, each experiment corresponds to a different set of hyperparameter configurations, denoted as HyperParameterk; where k represents the kth experiment, and the value range is k∈[1,j]

[0179] Step S302: pre-trained dataset A Pre-training Set the ratio to 8:2 and randomly divide it into training set A Pre-trainingtrain and test set A Pre-trainingtest , and are respectively brought into the BP neural network for forward propagation calculation;

[0180] S303: The BP neural network adopts the encoder-decoder architecture design. The encoder and decoder are respectively composed of several stacked transformer blocks. During pre-training, the training set X(i) is first positionally encoded through the encoder part, and then input into the encoder block of the transformer neural network for feature encoding, and finally the output data of the encoder is obtained;

[0181] S304: During pre-training, Y(i) is position-encoded and input into the decoder; the first transformer decoder block of the decoder network performs mask processing (MaskedSelf-Attention) on the input position-encoded target output data Y(i), and substitutes it and the encoder output data into the subsequent decoder block for training; the masked decoder input is interactively fused with the feature representation output by the encoder network, and the cross-attention mechanism is used to generate the target output sequence for training;

[0182] S305: Training is completed, and the final pre-trained BP neural network weight parameter configuration Pt is saved k , where k represents the kth weight parameter configuration.

[0183] Step S400: For the BP neural network pre-trained in S300, freeze the neural network topology and weights of all layers except the last layer, and obtain the initial weights of the BP neural network for transfer learning;

[0184] Step S500: fine-tune the BP neural network after transfer learning according to the target data set in S200 and the weights saved in S400.

[0185] Step S500: The target data set B collected on-site in S200 Fine-tuning Set the ratio to 8:2 and randomly divide it into training set B Fine-tuningtrain and test set B Fine-tuningtest ;

[0186] Step S5.2: Set the target training set B Fine-tuningtrain and test set B Fine-tuningtest According to the steps S3.1-S3.5, the training is input. The fine-tuning method only trains and modifies the hyperparameter configuration in the transfer learning framework. At the same time, the BP neural network weight parameter configuration after fine-tuning transfer learning is saved, which is recorded as Pt k ;

[0187] Step S502: Select the root mean square error (RMSE), mean absolute percentage error (MAPE), absolute percentage error (APE) and coefficient of determination (R 2)These four evaluation indicators evaluate the performance of the BP neural network model after transfer learning and are defined as follows:

[0188]

[0189] Among them, n means there are n prediction samples, y(i) and Respectively represent the actual value of the i-th sample and the predicted value of the model under n different working conditions; sort out RMSE, MAPE, APE and R 2 Save to error set C i , which can be expressed as:

[0190] C i =(RMSE i 、MAPE i 、APE、R 2 )

[0191] Step S600: Change the hidden layer of the pre-trained BP neural network, repeat S500, and find a group of neural networks with the best prediction effect, that is, the smallest loss function.

[0192] Step S601: Change the hyperparameter configuration in step S300, recorded as the hyperparameter configuration Hyperparameter i , where i∈[2,j]

[0193] Step S602: Repeat steps S3.3-S5.3 to obtain weight parameter configuration Pt in sequence. k , the fine-tuned weight parameter configuration Pt k , error set D k ;

[0194] Step S603: Repeat steps S6.1-S6.2 until the number of hyperparameter configuration debugging experiments reaches j times, and find the group with the smallest error in j experiments, which is recorded as D best , the corresponding fine-tuning parameter weight is recorded as Pt best , the hyperparameter configuration is recorded as Hyperparameter best , that is, the hyperparameter configuration with the best model prediction effect is obtained.

[0195] like Figure 1 As shown in steps S400-S600 and Table 2, the specific process of transfer learning in steps S400-S600 is as follows:

[0196] Table 2 Pseudo code of transfer learning based on BP neural network

[0197]

[0198] According to the calculation results, the optimal hyperparameter configuration HyperParameter best The settings are shown in Table 3:

[0199] Table 3 Optimal hyperparameter groups

[0200]

[0201]

[0202] Step S700: The BP neural network optimal prediction model based on transfer learning is used to predict the temperature at different positions in the incinerator under different working conditions.

[0203] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent prediction method for the temperature in an incinerator based on transfer learning, characterized in that: include, Obtain the temperature data of the numerically simulated incinerator to establish a pre-training data set, collect the actual incinerator operation data, and establish a target data set; Build a BP neural network model based on the pre-training data set, train the model using the pre-training data set, and save the weight parameters of the pre-training model; Freeze all layers of the pre-trained model except the last hidden layer, add a new hidden layer after the frozen layer, and fine-tune the newly added layer using the target dataset; Changing the number of hidden layers and hyperparameter configurations, repeating the transfer learning process, and selecting the model with the best prediction effect, wherein the model with the best prediction effect has the smallest loss function; Use the optimized transfer learning model to predict the temperature at different locations in the incinerator under different operating conditions.

2. The intelligent prediction method for the temperature in the incinerator based on transfer learning as claimed in claim 1, characterized in that: The method of obtaining the numerically simulated incinerator temperature data and establishing the pre-training data set includes: Obtain the temperature data of domestic waste and different positions in the furnace for numerical solution under different working conditions, and establish a data set for BP neural network pre-training; Collect temperature data of domestic waste and different positions in the incinerator under different working conditions during actual operation of the incinerator manufacturer, and establish a target dataset for transfer learning fine-tuning; Pre-train the simulation data set to obtain the initial weights of the BP neural network and save them for subsequent transfer learning; Freeze the neural network topology and weights of all layers except the last layer in the pre-trained model to obtain the initial BP neural network weights for transfer learning; Based on the target data set, migrate the topological structure and weight data saved by the frozen pre-trained model to fine-tune the topological structure of the BP neural network; Repeat the transfer learning process to find a set of neural networks that has the best prediction effect on the target data set, that is, the smallest loss function; The optimal prediction model of BP neural network based on transfer learning is used to predict the temperature at different positions in the incinerator under different working conditions.

3. The intelligent prediction method for the temperature in the incinerator based on transfer learning as claimed in claim 2 is characterized by: The method of obtaining the numerically solved domestic waste and temperature data at different positions in the furnace under different working conditions includes preprocessing the data obtained after pyrolysis and gasification of domestic waste in the incinerator under different working conditions as a data set for pre-training of the BP neural network, specifically including the following steps: Select factors that affect temperature change such as moisture content, volatile matter, fixed carbon, ash, garbage particle size, primary air, circulating air, secondary air, load, etc. of different domestic garbage under different operating parameters as input variables, and normalize all factors; Based on the average influence value algorithm, all influencing factors are reduced in dimension; The input of the BP neural network is expressed as: X(i)=[x(1),x(2),...,x(i),x(n)],i∈[1,n] x(i)=[x1(i),x2(i),...,xm(i)],i∈[1,n] Among them, X(i) represents the sequence of influencing factors under different working conditions i; x(i) is a vector composed of m different factors; the temperatures at different locations in the furnace used for prediction under different working conditions are as follows: Y(i)=[y(1),y(2),...,y(i),...,y(n)],i∈[1,n] Among them, Y(i) represents the temperature sequence of different positions in the furnace under different working conditions i, that is, for n different working conditions, each working condition i has an output vector y(i) composed of the temperatures at different positions in the furnace; For different working conditions i, the characteristic variables in x(i) are normalized, and the input features x(i) under each working condition are mapped to the interval [0,1] to eliminate dimensional differences and improve the stability and convergence of model training; The dataset used for training is constructed as follows: A={X(i),Y(i)}={(x(1),y(1)),(x(2),y(2)),(x(i),y(i)),…,(x(n),y(n))} Among them, (x(i), y(i)) represents the normalized data pair under the i-th working condition; Based on different working conditions, dataset A is further divided into pre-training dataset and prediction dataset.

4. The intelligent prediction method for the temperature in the incinerator based on transfer learning as claimed in claim 3 is characterized by: The on-site collection of temperature data of domestic garbage and different positions in the incinerator under different working conditions during the actual operation of an incinerator manufacturer includes repeating the following steps: The factors that affect the temperature change under different working conditions, such as the moisture content, volatile matter, fixed carbon, ash, garbage particle size, primary air, circulating air, secondary air, and load of different domestic garbage are selected as input variables, and all factors are normalized; Based on the average influence value algorithm, all influencing factors are reduced in dimension; The input of the BP neural network is expressed as: X(i)=[x(1),x(2),...,x(i),x(n)],i∈[1,n] x(i)=[x1(i),x2(i),...,xm(i)],i∈[1,n] Among them, X(i) represents the sequence of influencing factors under different working conditions i; x(i) is a vector composed of m different factors; The temperatures at different locations in the furnace used for prediction under different working conditions are as follows: Y(i)=[y(1),y(2),...,y(i),...,y(n)],i∈[1,n] Among them, Y(i) represents the temperature sequence of different positions in the furnace under different working conditions i, that is, for n different working conditions, each working condition i has an output vector y(i) composed of the temperatures at different positions in the furnace; For different working conditions i, the characteristic variables in x(i) are normalized, and the input features x(i) under each working condition are mapped to the interval [0,1] to eliminate dimensional differences and improve the stability and convergence of model training; The dataset used for training is constructed as follows: A={X(i),Y(i)}={(x(1),y(1)),(x(2),y(2)),(x(i),y(i)),…,(x(n),y(n))} Among them, (x(i), y(i)) represents the normalized data pair under the i-th working condition; Based on different working conditions, dataset B is further divided into a pre-training dataset and a prediction dataset.

5. The intelligent prediction method for the temperature in the incinerator based on transfer learning as claimed in claim 4 is characterized in that: The data set A is pre-trained to obtain the initialization weights of the BP neural network, which are saved for subsequent transfer learning. The initialization training adopts the following steps: According to the BP neural network architecture, the hyperparameter configuration HyperParameter is optimized, and j hyperparameter configuration experiments are set. Each experiment corresponds to a different set of hyperparameter configurations, recorded as HyperParameter k ; where k represents the kth experiment, and its value range is k∈[1,j]; The pre-trained data set is set to a ratio of 8:2, randomly divided into training set and test set, and brought into the BP neural network for forward propagation calculation respectively; The BP neural network adopts the encoder-decoder architecture design. The encoder and decoder are respectively composed of several stacked transformer blocks. During pre-training, the training set X(i) is first positionally encoded through the encoder part, and then input into the encoder block of the transformer neural network for feature encoding, and finally the output data of the encoder is obtained; During pre-training, Y(i) is positionally encoded and then input into the decoder; The first transformer decoder block of the decoder network masks the input position-encoded target output data Y(i) and substitutes it together with the encoder output data into the subsequent decoder block for training; The masked decoder input is interactively fused with the feature representation of the encoder network output, and the cross-attention mechanism is used to generate the target output sequence for training; Training is completed, save the final pre-trained BP neural network weight parameter configuration Pt k , where k represents the kth weight parameter configuration; The neural network topology and weights of all layers except the last layer in the pre-trained model are frozen to obtain the initial BP neural network weights for transfer learning, so that its weight parameters do not change during training, and the BP neural network after transfer learning is obtained.

6. The intelligent prediction method for the temperature in the incinerator based on transfer learning as claimed in claim 5, characterized in that: Based on the target data set, the topological structure and weight data saved by the migration and freezing pre-training model are fine-tuned to the topological structure of the BP neural network. The specific steps of fine-tuning are as follows: The target data set collected on site was set to a ratio of 8:2 and randomly divided into training and test sets; The target training set and test set are input into the training according to the initialization training method. The fine-tuning method only trains and modifies the hyperparameter configuration in the transfer learning architecture. At the same time, the BP neural network weight parameter configuration after fine-tuning transfer learning is saved, which is recorded as Pt k ′; The four evaluation indicators of root mean square error, mean absolute percentage error, absolute percentage error and determination coefficient are selected to evaluate the performance of the BP neural network model after transfer learning, which are defined as follows: Among them, RMSE represents the root mean square error, MAPE represents the mean absolute percentage error, APE represents the absolute percentage error and R 2 represents the coefficient of determination; n represents a total of n prediction samples, y(i) and They represent the actual value of the i-th sample and the predicted value of the model under n different working conditions respectively; RMSE, MAPE, APE and R2 are sorted and saved in the error set Ci, which can be expressed as: C i =(RMSE i 、MAPE i 、APE、R 2 ) Among them, RMSE represents the root mean square error, MAPE represents the mean absolute percentage error, APE represents the absolute percentage error and R 2 represents the coefficient of determination.

7. The intelligent prediction method of the temperature in the incinerator based on transfer learning as claimed in claim 6, characterized in that: The specific steps of repeating the transfer learning process to find a set of neural networks with the best prediction effect on the target data set, that is, the smallest loss function, are as follows: Change the hyperparameter configuration, recorded as the hyperparameter configuration HyperParameter i , where i∈[2,j]; Repeated BP neural network adopts the encoder-decoder architecture design-fine-tuning specific steps to obtain the weight parameter configuration Pt in turn k , the fine-tuned weight parameter configuration Pt k ′, error set D k ; Repeat the transfer learning process twice until the number of hyperparameter configuration debugging experiments reaches j times, and find the group with the smallest error in j experiments, which is recorded as D best , the corresponding fine-tuning parameter weight is recorded as Pt best ′, the hyperparameter configuration is recorded as HyperParameter best , that is, the hyperparameter configuration with the best model prediction effect is obtained; The effect of the BP neural network without transfer learning was compared with that of the BP neural network after transfer learning. The predicted values ​​and error sets obtained by the two algorithms were compared and analyzed. Finally, the optimal prediction model of transfer learning based on the BP neural network was obtained as a prediction method for the temperature at different positions in the domestic waste incinerator under different working conditions.

8. An intelligent prediction system for the temperature inside an incinerator based on transfer learning, based on the intelligent prediction method for the temperature inside an incinerator based on transfer learning according to any one of claims 1 to 7, characterized in that: It also includes an acquisition module, which acquires the temperature data of the incinerator of the numerical simulation, establishes a pre-training data set, collects the actual incinerator operation data, and establishes a target data set; The pre-training module builds a BP neural network model based on the pre-training data set, trains the model using the pre-training data set, and saves the weight parameters of the pre-training model; Freeze module, freeze all layers of the pre-trained model except the last hidden layer, add a new hidden layer after the frozen layer, and use the target dataset to fine-tune the newly added layer; Iterate the module, change the number of hidden layers and hyperparameter configuration, repeat the transfer learning process, and select the model with the best prediction effect, wherein the model with the best prediction effect has the smallest loss function; The prediction module uses the optimized transfer learning model to predict the temperature at different locations in the incinerator under different operating conditions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent prediction method of the temperature in the incinerator based on transfer learning as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent prediction method for the temperature inside the incinerator based on transfer learning as described in any one of claims 1 to 7 are implemented.

Images