A Deep Learning-Based Method for Predicting and Identifying Anomalies in the Thermal Treatment of Solid Waste

Abstract

This invention discloses a deep learning-based method for predicting and identifying anomalies in the thermal treatment of solid waste, belonging to the field of solid waste thermal treatment technology. It aims to address the problem of comprehensive monitoring and intelligent early warning of the solid waste thermal treatment process. The invention constructs a first hidden layer, a second hidden layer, and an output layer of a deep learning-based solid waste thermal treatment prediction model; designs a loss function for the model; and sequentially connects the input layer, first hidden layer, second hidden layer, and output layer to obtain the deep learning-based solid waste thermal treatment prediction model. Finally, it constructs an anomaly prediction method for solid waste thermal treatment, analyzing anomalies from the perspectives of material transformation anomalies, material distribution anomalies, and energy transformation anomalies, and providing an anomaly diagnosis and judgment scheme.

Description

Technical Field

[0001] This invention belongs to the field of solid waste thermal treatment technology, specifically involving a method for predicting and anomaly detection in solid waste thermal treatment based on deep learning. Background Technology

[0002] The thermal treatment of solid waste involves complex transformations of matter and energy. First, the composition of solid waste is highly heterogeneous and uncertain, including various components such as combustibles, moisture, and inorganic substances, whose proportions change continuously with time and source. Second, multiple physicochemical processes, including drying, pyrolysis, and combustion, occur simultaneously during thermal treatment, exhibiting significant coupling effects. The drying process influences subsequent pyrolysis and combustion efficiency, the composition of pyrolysis products affects combustion characteristics, and the heat released during combustion, in turn, affects drying and pyrolysis. Simultaneously, the distributions of temperature, flow, and concentration fields also interact, forming complex multi-field couplings. Furthermore, multi-scale processes such as heat and mass transfer, chemical reactions, and phase transitions occur within the furnace; the nonlinear characteristics and dynamic changes of these processes pose significant challenges to process control.

[0003] In the thermal treatment of solid waste, conversion efficiency directly reflects the degree of material conversion, affecting the completeness of incineration and the generation of pollutants. For example, carbon conversion rate affects combustion completeness and carbon dioxide emissions, while sulfur and chlorine conversion rate are closely related to the generation of acidic gases. Product distribution determines pollution control strategies and resource utilization directions; accurate prediction helps optimize subsequent treatment processes. Energy conversion reflects the level of energy utilization and is crucial for improving system efficiency and reducing operating costs. Accurate prediction of these indicators can provide a basis for decision-making in process optimization and intelligent control, which is of great significance for improving system operating efficiency, reducing pollutant emissions, and ensuring safe and stable operation.

[0004] Current research on predicting key indicators in the thermal treatment of solid waste suffers from the following shortcomings: First, the prediction accuracy is generally low, especially when operating conditions fluctuate significantly, with predicted results deviating markedly from actual values. Second, existing models consider relatively singular influencing factors, often focusing only on certain key process parameters while neglecting the impact of important factors such as material properties and environmental conditions, resulting in poor adaptability to complex operating conditions. Furthermore, most models only predict one or a few indicators, lacking the comprehensive predictive capability for multiple key indicators such as conversion efficiency, product distribution, and energy conversion. There is an urgent need to establish a comprehensive model that considers more factors and provides better predictive results to meet practical engineering needs.

[0005] On the other hand, in the operation and management of solid waste thermal treatment processes, the complexity and uncertainty of the processes make timely detection and prediction of operational anomalies a persistent research challenge. Existing anomaly diagnosis methods often focus only on fluctuations in a single indicator, lacking a systematic assessment of multi-dimensional anomalies such as material transformation, product distribution, and energy conversion, making it difficult to accurately reflect the overall operating status of the system. Therefore, there is an urgent need to establish a scientific method for predicting anomalies in solid waste thermal treatment, enabling comprehensive monitoring and intelligent early warning of the solid waste thermal treatment process, providing a reliable basis for process adjustment and fault prevention. Summary of the Invention

[0006] The problem this invention aims to solve is to achieve comprehensive monitoring and intelligent early warning of the thermal treatment process of solid waste, and proposes a method for predicting and judging anomalies in the thermal treatment of solid waste based on deep learning.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A deep learning-based method for predicting and identifying anomalies in the thermal treatment of solid waste includes the following steps:

[0009] S1. Determine the input parameters of the input layer of the deep learning-based solid waste thermal treatment prediction model, including material characteristic parameters, process operation parameters, and process parameters;

[0010] S2. Construct the first hidden layer of a deep learning-based prediction model for the thermal treatment of solid waste;

[0011] S3. Construct the second hidden layer of a deep learning-based prediction model for the thermal treatment of solid waste;

[0012] S4. Construct the output layer of the deep learning-based solid waste thermal treatment prediction model, divide the output layer into conversion efficiency group, product distribution group and energy conversion group, and determine the output parameters of the output layer of the deep learning-based solid waste thermal treatment prediction model.

[0013] S5. Design the loss function for a deep learning-based prediction model of solid waste thermal treatment;

[0014] S6. Connect the input layer, first hidden layer, second hidden layer and output layer of the deep learning-based solid waste thermal treatment prediction model in sequence to obtain the deep learning-based solid waste thermal treatment prediction model.

[0015] S7. Collect data through the solid waste treatment plant and automated control system, normalize the data, divide the data into training set, validation set and test set in a ratio of 7:2:1, train, validate and test the deep learning-based solid waste thermal treatment prediction model obtained in step S6, and obtain the trained deep learning-based solid waste thermal treatment prediction model.

[0016] S8. After inverse normalization of the prediction results of the deep learning-based solid waste thermal treatment prediction model, the prediction results of the deep learning-based solid waste thermal treatment prediction model with actual physical quantities are obtained. An anomaly prediction method for solid waste thermal treatment is constructed. When considering the anomaly problem of solid waste thermal treatment, the analysis is carried out from the perspectives of abnormal material transformation, abnormal material distribution, and abnormal energy transformation, and an anomaly diagnosis and judgment scheme is given.

[0017] Furthermore, in step S1, the number of nodes in the input layer is determined to be 35, denoted as X1-X2. 35 ;

[0018] X1~X 14 These are material characteristic parameters, with the specific parameters as follows: X1 is moisture content, X2 is ash content, X3 is combustible content, X4 is carbon content, X5 is hydrogen content, X6 is oxygen content, X7 is nitrogen content, X8 is sulfur content, X9 is chlorine content, X... 10 X represents the calorific value. 11 X is the bulk density. 12 For particle size, X 13 X represents the sand content. 14 Metal content;

[0019] X 15 ~X 25 These are the process operation parameters, and the specific parameters are as follows: X 15 X is the grate speed. 16 X is the feed rate. 17 For primary air volume, X 18 For secondary air volume, X 19 X represents the furnace temperature. 20 For negative pressure in the furnace, X 21 X represents the flue temperature. 22 X represents the steam temperature. 23 X is the steam pressure. 24 X represents the steam flow rate. 25 The ambient temperature;

[0020] X 26 ~X 35 These are process parameters, and the specific parameters are as follows: X 26 X represents the feed moisture content. 27X represents the discharge moisture content. 28 For feed volatiles, X 29 For the discharge volatile matter, X 30 X is the feed mass flow rate. 31 X represents the discharge mass flow rate. 32 X is the calorific value of the feed. 33 X is the discharge calorific value. 34 X represents the oxygen concentration in the smoke. 35 This refers to the ash content.

[0021] Furthermore, step S2 divides the nodes of the first hidden layer into a material characteristic group, a reaction characteristic group, and an energy characteristic group. The material characteristic group focuses on the changes in the physicochemical properties of materials during heat treatment and contains 30 neuron nodes, with the i1th neuron node having a value of 1 to 30. The reaction characteristic group focuses on describing the extent of various chemical reactions and contains 25 neuron nodes, with the i2th neuron node having a value of 31 to 55. The energy characteristic group focuses on characterizing energy conversion and transfer features and contains 25 neuron nodes, with the i3th neuron node having a value of 56 to 80.

[0022] The specific implementation method of step S2 includes the following steps:

[0023] S2.1. Temperature correction and material moisture correction are introduced into the material property group. Temperature correction considers the driving effect of temperature on the material conversion rate and reflects the differences in the material conversion law in different temperature ranges; material moisture correction considers the influence of the moisture evaporation process on subsequent conversion.

[0024] S2.2. Constructing a reaction characteristic group introduces a mass conversion term to directly reflect the degree of reaction of substances and embody the depth of the reaction; adding an excess air term characterizes the sufficiency of the combustion environment and reflects the influence of oxidation reaction conditions on the conversion effect;

[0025] S2.3. Construct an energy characteristic group by introducing a calorific value conversion term to directly characterize the energy conversion effect; add a temperature gradient term to reflect the driving force and intensity of heat transfer, and describe the spatial distribution pattern of energy.

[0026] Furthermore, step S3 divides the nodes of the second hidden layer into a drying process group, a pyrolysis process group, and a combustion process group. The drying process group describes the influence of the drying process on subsequent pyrolysis and combustion, and contains 15 neuron nodes, with the i4th neuron node having a value of 1 to 15. The pyrolysis process group describes the compositional changes of the pyrolysis products, and contains 20 neuron nodes, with the i5th neuron node having a value of 16 to 35. The combustion process group describes the process characteristics of the final oxidation stage, and contains 15 neuron nodes, with the i6th neuron node having a value of 36 to 50.

[0027] The specific implementation method of step S3 includes the following steps:

[0028] S3.1. Constructing a drying process group: By characterizing the process features of moisture removal from solid waste, reflecting the driving effect of temperature on moisture evaporation, and describing the impact of the drying process on subsequent pyrolysis and combustion;

[0029] S3.2. Constructing a pyrolysis process group: By reflecting the kinetic characteristics of organic matter pyrolysis and decomposition, the influence of temperature on volatile matter is analyzed, the compositional changes of pyrolysis products are described, and key characteristics of the pyrolysis stage are extracted.

[0030] S3.3. Construct a combustion process group to characterize the oxidation process of fixed carbon and volatiles, reflect the effect of oxygen concentration on combustion efficiency, and describe the process characteristics of the final oxidation stage.

[0031] Furthermore, step S4, based on the evaluation requirements of the solid waste thermal treatment process, divides the output layer into a conversion efficiency group, a product distribution group, and an energy conversion group; the conversion efficiency group indicators reflect the completeness and sufficiency of material conversion, the product distribution group indicators reveal the migration patterns of materials and environmental impacts, and the energy conversion group indicators reflect the energy utilization level.

[0032] The conversion efficiency group indicators include Y1~Y6, where Y1 is the moisture conversion rate, Y2 is the volatile matter conversion rate, Y3 is the fixed carbon conversion rate, Y4 is the carbon conversion rate, Y5 is the sulfur conversion rate, and Y6 is the chlorine conversion rate.

[0033] Product distribution group indicators include the material distribution ratio indicator Y. 71 ~ Y 75 And residue allocation index Y8~Y9, Y 71 Y represents the mass distribution ratio of moisture entering the flue gas after drying. 72 Y represents the mass distribution ratio of volatile matter entering the flue gas after pyrolysis. 73 Y represents the proportion of carbon elements converted into carbon dioxide that enters the flue gas. 74 Y represents the proportion of sulfur that has been converted into sulfur dioxide and enters the flue gas. 75 Y8 represents the mass distribution ratio of chlorine after it is converted into hydrogen chloride and enters the flue gas; Y9 represents the mass distribution ratio of fixed carbon residue in the slag after combustion; and Y9 represents the mass distribution ratio of fly ash components after the combustion of fixed carbon and enters the fly ash.

[0034] Energy conversion group indicators include Y 10 ~Y 12 Y 10 For thermal efficiency, Y 11 For combustion efficiency, Y 12 This represents the heat loss rate.

[0035] Furthermore, the loss function in step S5 considers both prediction loss and heat treatment loss. The specific implementation method includes the following steps:

[0036] S5.1. Construct the prediction loss based on the conversion efficiency prediction loss, product distribution prediction loss, and energy conversion prediction loss, with the following expression:

[0037]

[0038] in, Predicting losses for conversion efficiency; Predicting losses for product distribution; Energy conversion prediction loss;

[0039]

[0040] in, This is the function for calculating the mean square error; Output values ​​for different indicators in the conversion efficiency group predicted by the model; Output values ​​for different indicators in the actual measured conversion efficiency group;

[0041]

[0042] in, Output values ​​for different indicators in the product distribution group predicted by the model; Output values ​​for different indicators in the actual measured product distribution group;

[0043]

[0044] in, Output values ​​for different indicators in the energy conversion group predicted by the model; These are the output values ​​of different indicators in the actual measured energy conversion group;

[0045] S5.2. Constructing heat treatment loss based on material conversion constraints, composition conversion constraints, and heat conversion constraints. The expression is:

[0046]

[0047] in, Material conversion constraints The weights corresponding to material conversion constraints; Constraints on component transformation The weights corresponding to the component transformation constraints; For heat conversion constraints, The weights corresponding to the heat conversion constraints;

[0048] Material conversion constraints are established as follows:

[0049]

[0050] in, For volatile matter conversion terms; For fixed carbon allocation items;

[0051] The component transformation constraint is established as follows:

[0052]

[0053] in, Constraints on sulfur element transformation Constraints for chlorine element transformation;

[0054] The heat conversion constraint is established as follows:

[0055] ;

[0056] S5.3. Construct the total loss based on predicted loss and heat treatment loss. ;

[0057]

[0058] in, The weights corresponding to the predicted loss, The weights corresponding to heat treatment losses are determined by expert experience.

[0059] Furthermore, in step S7, the model training uses the backpropagation algorithm. During training, the model calculates the predicted value through forward propagation, substitutes the predicted value and the true value into the loss function to calculate the loss, and then updates the network parameters through backpropagation. To prevent overfitting, an early stopping strategy is adopted, and training is stopped when the validation set loss does not decrease for 10 consecutive epochs. The output result is restored to the actual physical quantity through inverse normalization.

[0060] Furthermore, the specific implementation method of step S8 includes the following steps:

[0061] S8.1. In the process of identifying anomalies in the thermal treatment of solid waste, the degree of material transformation anomaly reflects the degree of independent deviation of each process parameter, which helps to accurately locate the abnormal link and take targeted adjustments. The calculation formula is:

[0062]

[0063] in, Output values ​​for different indicators predicted by the model. include , , ; To and The corresponding standard values ​​are determined by expert experience, design documents, and experimental research methods; The allowable fluctuation range corresponding to the formula for the degree of material transformation anomaly is determined by expert experience, design documents, experimental research, and other methods. for and The corresponding allocation weight coefficients;

[0064] S8.2. The anomaly degree of material distribution includes the anomaly degree of flue gas distribution and the anomaly degree of solid distribution;

[0065] Considering the distribution deviations of each component in the flue gas during moisture gasification, volatile matter desorption, carbon oxidation, sulfur oxidation, and hydrogen chloride generation, calculate the flue gas distribution anomaly degree. The expression is:

[0066]

[0067] in, To assign weighting coefficients, s2 = 71, 72, 73, 74, 75; To and The corresponding standard values ​​are determined by expert experience, design documents, experimental research, and other methods. This represents the allowable fluctuation range corresponding to the flue gas distribution anomaly formula.

[0068] Considering the distribution equilibrium of inorganic matter in slag and fly ash, and combining the influence of process parameters such as combustion temperature, residence time, and gas-solid separation efficiency on the slag-to-fly ash ratio, the degree of solid distribution anomaly is calculated. The expression is:

[0069]

[0070] in, , The weighting coefficients are assigned to slag and fly ash, respectively. , These represent the allowable fluctuation ranges corresponding to the weighting coefficients of slag and fly ash in the solid distribution anomaly formula;

[0071] S8.3. Energy conversion anomaly degree includes energy anomaly degree. and energy balance anomaly ;

[0072] Considering the impact of factors such as solid waste calorific value fluctuations, combustion air distribution, furnace temperature distribution, and heat exchange efficiency on system thermal efficiency, combustion efficiency, and heat loss, the energy anomaly degree is calculated. The expression is:

[0073]

[0074] in, For the energy index weights, s3 = 10, 11, 12; This represents the allowable fluctuation range corresponding to the energy anomaly formula;

[0075] Considering the balance between effective heat output and heat loss in the system, and taking into account factors such as furnace wall insulation performance and cooling system efficiency, the energy balance anomaly degree is calculated. The expression is:

[0076]

[0077] in, The allowable error for energy balance is determined by expert experience, design documents, experimental research, and other methods.

[0078] S8.4. Construct the comprehensive anomaly index ZH;

[0079]

[0080] in, , , , , They are respectively with , , , , The corresponding process weighting coefficients are determined by expert experience, design documents, experimental research, and other methods.

[0081] S8.5. Based on the comprehensive abnormality index obtained in step S4, perform anomaly diagnosis and judgment;

[0082] The anomaly levels are divided into four categories: normal operation, minor anomaly, moderate anomaly, and severe anomaly. The critical comprehensive anomaly indices corresponding to minor, moderate, and severe anomalies are as follows: , , Then we get For normal operation; It is a minor abnormality; Moderately abnormal; This is a serious abnormality.

[0083] The beneficial effects of this invention are:

[0084] This invention discloses a deep learning-based method for predicting and identifying anomalies in the thermal treatment of solid waste. The analysis of the solid waste thermal treatment process considers conversion efficiency, product distribution, and energy conversion. Conversion efficiency reflects the completeness and sufficiency of material conversion; product distribution reveals the migration patterns and environmental impacts of materials; and energy conversion reflects the level of energy utilization. These three aspects constitute a complete characterization system for the solid waste thermal treatment process, comprehensively reflecting the operational effectiveness of the process and providing important basis for process optimization and control.

[0085] This invention discloses a deep learning-based method for predicting and identifying anomalies in the thermal treatment of solid waste. The input parameters, first hidden layer, second hidden layer, and output layer are all specifically designed and optimized to address the characteristics of solid waste thermal treatment. This achieves scientific and reasonable predictions of conversion efficiency, product distribution, and energy conversion during the solid waste thermal treatment process. Using the established deep learning model, anomalies in the state of solid waste thermal treatment are determined from the perspectives of material conversion anomalies, material distribution anomalies, and energy conversion anomalies. This allows for a comprehensive evaluation of system performance, timely detection and handling of potential operational problems, and ensures the continuity and safety of the solid waste treatment process. Attached Figure Description

[0086] Figure 1 This is a flowchart of a method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning, as described in this invention.

[0087] Figure 2 This is a loss curve diagram of the training process of the solid waste thermal treatment prediction model of the present invention;

[0088] Figure 3 This is a comparison chart of the predicted values ​​and actual values ​​of the solid waste thermal treatment prediction model of this invention;

[0089] Figure 4 This is a graph showing the calculation results of the comprehensive anomaly index of this invention. Detailed Implementation

[0090] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described specific embodiments are merely a part of the embodiments of the invention, and not all of them. The components of the specific embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations, and the invention may also have other embodiments.

[0091] Therefore, the following detailed description of specific embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected specific embodiments of the invention. All other specific embodiments obtained by those skilled in the art based on these specific embodiments without inventive effort are within the scope of protection of this invention.

[0092] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figure 1 -Appendix Figure 4 Detailed explanation is as follows:

[0093] Example 1:

[0094] A deep learning-based method for predicting and identifying anomalies in the thermal treatment of solid waste includes the following steps:

[0095] S1. Determine the input parameters of the input layer of the deep learning-based solid waste thermal treatment prediction model, including material characteristic parameters, process operation parameters, and process parameters;

[0096] Material characteristic parameters reflect the basic properties of solid waste, including composition (moisture, ash, combustibles, etc.), elemental composition (C, H, O, N, S, Cl), and physical properties (calorific value, bulk density, etc.), which directly affect the material conversion process. Process operating parameters include controllable operating conditions such as grate speed, air volume, temperature, and pressure, which determine the reaction environment. Process parameters reflect the dynamic changes in matter and energy during the conversion process, such as the moisture content, volatile matter, mass flow rate, and calorific value of the feed and discharge. This classification method comprehensively considers the key factors affecting process characteristics, helping the model establish the intrinsic relationship between material properties, operating conditions, and process characteristics.

[0097] Furthermore, in step S1, the number of nodes in the input layer is determined to be 35, denoted as X1-X2. 35 ;

[0098] X1~X 14 These are material characteristic parameters, with the specific parameters as follows: X1 is moisture content, X2 is ash content, X3 is combustible content, X4 is carbon content, X5 is hydrogen content, X6 is oxygen content, X7 is nitrogen content, X8 is sulfur content, X9 is chlorine content, X... 10 X represents the calorific value. 11 X is the bulk density. 12 For particle size, X 13 X represents the sand content. 14 Metal content;

[0099] X 15 ~X 25These are the process operation parameters, and the specific parameters are as follows: X 15 X is the grate speed. 16 X is the feed rate. 17 For primary air volume, X 18 For secondary air volume, X 19 X represents the furnace temperature. 20 For negative pressure in the furnace, X 21 X represents the flue temperature. 22 X represents the steam temperature. 23 X is the steam pressure. 24 X represents the steam flow rate. 25 The ambient temperature;

[0100] X 26 ~X 35 These are process parameters, and the specific parameters are as follows: X 26 X represents the feed moisture content. 27 X represents the discharge moisture content. 28 For feed volatiles, X 29 For the discharge volatile matter, X 30 X is the feed mass flow rate. 31 X represents the discharge mass flow rate. 32 X is the calorific value of the feed. 33 X is the discharge calorific value. 34 X represents the oxygen concentration in the smoke. 35 This refers to the ash content.

[0101] All of the above parameters can be directly collected through an automated control system. The data comes from historical data from a solid waste treatment plant. To ensure the training effect and prediction accuracy of the deep learning model, normalization of the input and output parameters is necessary. Normalization eliminates the influence of different units on the model, ensuring all features are within the same scale, which helps accelerate model convergence and improve model performance. Simultaneously, normalization avoids numerical overflow or underflow problems during numerical calculations. For the element content indices X1~X9 in the material characteristic parameters, since they are all mass percentages, they are directly divided by 100 to map to the 0-1 range; X 10 ~ X 12 Then a linear mapping is performed based on the maximum and minimum value range of historical data statistics; X 13 and X 14 As a percentage, directly divide by 100 for mapping. In the process operating parameters, X... 15 and X 16 Mapping is performed based on the device's rated range, X 17 and X 18 Based on the rated range mapping of the blower, X 19 ~ X 25All are linearly mapped based on the statistical range of historical operating data. X in the process parameters... 26 ~ X 29 As a percentage, X is directly divided by 100 and mapped. 30 X 31 Based on the design operating condition range mapping, X 32 X 33 Based on historical data range mapping, X 34 Based on the theoretical air excess coefficient range mapping, X 35 As a percentage, it is directly divided by 100 for mapping. Through this normalization process, all parameters are uniformly mapped to the 0-1 range, laying the foundation for the subsequent training and prediction of deep learning models.

[0102] S2. Construct the first hidden layer of a deep learning-based prediction model for the thermal treatment of solid waste;

[0103] Furthermore, step S2 divides the nodes of the first hidden layer into a material characteristic group, a reaction characteristic group, and an energy characteristic group. The material characteristic group focuses on the changes in the physicochemical properties of materials during heat treatment and contains 30 neuron nodes, with the i1th neuron node having a value of 1 to 30. The reaction characteristic group focuses on describing the extent of various chemical reactions and contains 25 neuron nodes, with the i2th neuron node having a value of 31 to 55. The energy characteristic group focuses on characterizing energy conversion and transfer features and contains 25 neuron nodes, with the i3th neuron node having a value of 56 to 80.

[0104] The specific implementation method of step S2 includes the following steps:

[0105] S2.1. Temperature correction and material moisture correction are introduced into the material property group. Temperature correction considers the driving effect of temperature on the material conversion rate and reflects the differences in the material conversion law within different temperature ranges; material moisture correction considers the influence of moisture evaporation on subsequent conversions; the calculation method of the 30 neuron nodes in the material property group is as follows:

[0106]

[0107] in, This represents the output value of the i1th neuron node in the first hidden layer, where i1 ranges from 1 to 30. This is an activation function used to introduce nonlinear features; The weight coefficients are the weights of the i1th neuron node in the first hidden layer corresponding to the jth input parameter. For input parameters, ; This is the bias term for the i1th neuron node in the first hidden layer corresponding to the jth input parameter; The temperature correction factor is determined through expert experience, design documents, experimental research, and other methods. The moisture correction factor is determined by expert experience, design documents, experimental research, and other methods. This is a temperature correction term; This is a correction term for material moisture content; however, traditional deep learning methods only extract features through weights and biases, lacking a mechanistic representation of the material transformation process. This simple mathematical mapping is insufficient to accurately describe the dynamic changes in material properties with temperature, nor can it reflect the material transformation characteristics in the pretreatment stage.

[0108] S2.2. Constructing the reaction characteristic group introduces a mass conversion term to directly reflect the degree of reaction and the depth of the reaction; adding an excess air term characterizes the sufficiency of the combustion environment and reflects the influence of oxidation reaction conditions on the conversion effect; the calculation method of the 25 neuron nodes in the reaction characteristic group is as follows:

[0109]

[0110] in, This is the output value of the i2th neuron node in the first hidden layer, where i2 takes values ​​from 31 to 55; The weight coefficients are the weights of the i2-th neuron node in the first hidden layer corresponding to the j-th input parameter. This is the bias term for the i2-th neuron node in the first hidden layer corresponding to the j-th input parameter; The quality conversion coefficient is determined by expert experience, design documents, experimental research, and other methods. The excess air coefficient is determined through expert experience, design documents, experimental research, and other methods. The density of air is determined by actual measurement. For quality conversion items; The excess air term is used; however, traditional neural network models provide overly simplistic descriptions of the reaction process, relying solely on data and lacking consideration of the chemical reaction mechanism. The models cannot accurately reflect the influence of reaction conditions on the conversion process, leading to unreliable predictions. This improvement in this embodiment enhances the model's ability to perceive reaction kinetics and environmental factors.

[0111] S2.3. Constructing an energy characteristic group introduces a calorific value conversion term to directly characterize the energy conversion effect; a temperature gradient term is added to reflect the driving force and intensity of heat transfer, describing the spatial distribution of energy; the 25 neuron nodes in the energy characteristic group are calculated as follows:

[0112]

[0113] in, This is the output value of the i3th neuron node in the first hidden layer, where i3 takes values ​​from 56 to 80; The weight coefficients of the j-th input parameter and the corresponding i3-th neuron node in the first hidden layer; This is the bias term for the i3rd neuron node in the first hidden layer corresponding to the jth input parameter; This is a calorific value conversion term; The calorific value conversion coefficient is determined by expert experience, design documents, experimental research, and other methods. This is the temperature gradient term; The temperature gradient coefficient can be determined through expert experience, design documents, experimental research, etc. Conventional deep learning models often neglect the fundamental physical laws of energy transfer and conversion when dealing with energy conversion problems, leading to predictions that may violate the law of conservation of energy. This embodiment's improvement enables the model to simultaneously consider energy conversion and transfer characteristics, thus improving the accuracy of predictions regarding energy utilization processes.

[0114] Step S3 divides the nodes of the second hidden layer into a drying process group, a pyrolysis process group, and a combustion process group. The drying process group describes the influence of the drying process on subsequent pyrolysis and combustion, and contains 15 neuron nodes, with the i4th neuron node having a value of 1 to 15. The pyrolysis process group describes the compositional changes of the pyrolysis products, and contains 20 neuron nodes, with the i5th neuron node having a value of 16 to 35. The combustion process group describes the process characteristics of the final oxidation stage, and contains 15 neuron nodes, with the i6th neuron node having a value of 36 to 50.

[0115] S3. Construct a second hidden layer for a deep learning-based prediction model of solid waste thermal treatment; this design is based on the fundamental reaction mechanism of solid waste thermal treatment, reflecting the sequence and mutual influence of material transformation.

[0116] The specific implementation method of step S3 includes the following steps:

[0117] S3.1. The drying process group is constructed to characterize the process features of moisture removal from solid waste, reflect the driving effect of temperature on moisture evaporation, and describe the impact of the drying process on subsequent pyrolysis and combustion. The 15 neuron nodes in the drying process group are calculated as follows:

[0118]

[0119] in, This is the output value of the i4th neuron node in the second hidden layer, where i4 takes a value from 1 to 15; These are the weight coefficients of the i4th neuron node in the second hidden layer, corresponding to the kth neuron in the first hidden layer. =1~80; This is the bias term for the i4th neuron node in the second hidden layer, which corresponds to the kth neuron in the first hidden layer; This represents the output value of the k-th neuron in the first hidden layer. This is the drying process item. This is the drying process coefficient, determined by expert experience, design documents, experimental research, and other methods.

[0120] S3.2. Constructing a pyrolysis process group: By analyzing the kinetic characteristics of organic matter pyrolysis, this group reflects the influence of temperature on volatile matter, describes the compositional changes of pyrolysis products, and extracts key features of the pyrolysis stage. The calculation method for the 20 neuron nodes in the pyrolysis process group is as follows:

[0121]

[0122] in, This is the output value of the i5th neuron node in the second hidden layer, where i5 ranges from 16 to 35. The weight coefficients are the weights of the i5th neuron node in the second hidden layer, corresponding to the kth neuron in the first hidden layer. This is the bias term for the i5th neuron node in the second hidden layer, which corresponds to the kth neuron in the first hidden layer; This is the pyrolysis process term. The coefficient for the pyrolysis process is determined by expert experience, design documents, experimental research, and other methods.

[0123] Furthermore, it enables the extraction of key features of the pyrolysis stage.

[0124] S3.3. Construct a combustion process group to characterize the oxidation process of fixed carbon and volatiles, reflect the effect of oxygen concentration on combustion efficiency, and describe the process characteristics of the final oxidation stage; the calculation method of the 15 neuron nodes in the combustion process group is as follows:

[0125]

[0126] in, This is the output value of the i6th neuron node in the second hidden layer, where i6 takes a value of 36 to 50; The weight coefficients are the weights of the i6th neuron node in the second hidden layer, corresponding to the kth neuron in the first hidden layer. This is the bias term for the i6th neuron node in the second hidden layer, which corresponds to the kth neuron in the first hidden layer; For the combustion process, The combustion process coefficient, The reference oxygen concentration is determined by expert experience, design documents, and experimental research methods.

[0127] S4. Construct the output layer of the deep learning-based solid waste thermal treatment prediction model, divide the output layer into conversion efficiency group, product distribution group and energy conversion group, and determine the output parameters of the output layer of the deep learning-based solid waste thermal treatment prediction model.

[0128] Furthermore, step S4, based on the evaluation requirements of the solid waste thermal treatment process, divides the output layer into a conversion efficiency group, a product distribution group, and an energy conversion group; the conversion efficiency group indicators reflect the completeness and sufficiency of material conversion, the product distribution group indicators reveal the migration patterns of materials and environmental impacts, and the energy conversion group indicators reflect the energy utilization level.

[0129] The conversion efficiency group indicators include Y1~Y6, where Y1 is the moisture conversion rate, Y2 is the volatile matter conversion rate, Y3 is the fixed carbon conversion rate, Y4 is the carbon conversion rate, Y5 is the sulfur conversion rate, and Y6 is the chlorine conversion rate.

[0130] Product distribution group indicators include the material distribution ratio indicator Y. 71 ~ Y 75 And residue allocation index Y8~Y9, Y 71 Y represents the mass distribution ratio of moisture entering the flue gas after drying. 72 Y represents the mass distribution ratio of volatile matter entering the flue gas after pyrolysis. 73 Y represents the proportion of carbon elements converted into carbon dioxide that enters the flue gas. 74 Y represents the proportion of sulfur that has been converted into sulfur dioxide and enters the flue gas. 75 Y8 represents the mass distribution ratio of chlorine after it is converted into hydrogen chloride and enters the flue gas; Y9 represents the mass distribution ratio of fixed carbon residue in the slag after combustion; and Y9 represents the mass distribution ratio of fly ash components after the combustion of fixed carbon and enters the fly ash.

[0131] Energy conversion group indicators include Y 10 ~Y 12 Y 10 For thermal efficiency, Y 11 For combustion efficiency, Y 12 This refers to the heat loss rate;

[0132] For output parameters, a normalization method must first be used to map and transform the data. Conversion rate indicators (Y1-Y6) are inherently in the range of 0-1; material distribution ratio indicators (Y... 71 -Y 75 The percentage values ​​for ) and residue allocation indicators (Y8-Y9) are naturally limited to the range of 0-1; efficiency indicators (Y 10 -Y 12After determining a reasonable range based on actual system operating data, normalization mapping is performed. Through this normalization process, all parameters are uniformly mapped to the same numerical range, laying the foundation for subsequent training and prediction of deep learning models.

[0133] The specific implementation method of step S4 includes the following steps:

[0134] S4.1. The calculation method for the output values ​​of different indicators in the conversion efficiency group is as follows:

[0135]

[0136] in, s1 represents the output values ​​of different indicators in the conversion efficiency group; , where is the weighting coefficient of the output values ​​of different indicators in the conversion efficiency group corresponding to the v-th neuron in the second hidden layer; This is a bias term for the output values ​​of different indicators in the conversion efficiency group corresponding to the v-th neuron in the second hidden layer; This represents the output value of the v-th neuron in the second hidden layer.

[0137] When s1=1, v∈{1,2,…,15}; when s1=2, v∈{16,17,…,35}; when s1=3~6, v∈{36,37,…,50}.

[0138] Furthermore, the conversion efficiency groups reflect the transformation characteristics of the most critical components in solid waste during the thermal treatment process: moisture conversion rate characterizes the drying effect, volatile matter and fixed carbon conversion rates reflect the degree of pyrolysis and combustion, and carbon, sulfur, and chlorine conversion rates reflect the transformation patterns of potential pollutants, comprehensively reflecting the material transformation effect. The moisture conversion rate Y1 is associated with the drying process group because moisture evaporation mainly occurs during the drying stage; the volatile matter conversion rate Y2 is associated with the pyrolysis process group because volatile matter release occurs during the pyrolysis stage; and the fixed carbon, carbon, sulfur, and nitrogen conversion rates (Y3-Y6) are associated with the combustion process group because the transformation of these elements is mainly completed during the combustion stage.

[0139] S4.2. The calculation method for the output values ​​of different indicators in the product distribution group is as follows:

[0140]

[0141] in, s1 represents the output values ​​of different indicators in the product distribution group, and s2 represents different indicators in the product distribution group; , where is the weighting coefficient of different index output values ​​in the product distribution group corresponding to the v-th neuron in the second hidden layer; This is the bias term for the output values ​​of different indicators in the product distribution group corresponding to the v-th neuron in the second hidden layer;

[0142] When s2=71, v∈{1,2,…,15}; when s2=72, v∈{16,17,…,35}; when s2=73,74,75,8,9, v∈{36,37,…,50}.

[0143] The product distribution group selected three component distribution indicators: flue gas, slag, and fly ash. The flue gas component distribution includes: the mass distribution ratio of moisture entering the flue gas after drying (Y71); the mass distribution ratio of volatile matter entering the flue gas after pyrolysis (Y72); the distribution ratio of carbon entering the flue gas after conversion to carbon dioxide (Y73); the distribution ratio of sulfur entering the flue gas after conversion to sulfur dioxide (Y74); and the distribution ratio of chlorine entering the flue gas after conversion to hydrogen chloride (Y75). The slag component distribution is the mass distribution ratio of fixed carbon remaining in the slag after combustion (Y8). The fly ash component distribution is the mass distribution ratio of fixed carbon entering the fly ash after combustion (Y9). These indicators not only meet material balance requirements but also reflect the influence of process parameters on product distribution. Furthermore, they are closely related to environmental impact control, and these three indicators can comprehensively characterize the migration patterns of substances in different phases. The proportion of moisture in the flue gas, Y71, comes from the drying stage, so it is associated with the drying process group; the proportion of volatile matter entering the flue gas, Y72, originates from the pyrolysis stage, so it is associated with the pyrolysis process group; while the proportions of flue gas after the conversion of carbon, sulfur, and nitrogen elements (Y73-Y75) and the proportions of slag and fly ash (Y8-Y9) are all generated in the combustion stage, so they are associated with the combustion process group.

[0144] S4.3. The calculation method for the output values ​​of different indicators in the energy conversion group is as follows:

[0145]

[0146] in, s1 represents the output value of different indicators in the energy conversion group, and s2 represents different indicators in the energy conversion group. , where is the weighting coefficient of different index output values ​​in the energy conversion group corresponding to the v-th neuron in the second hidden layer; This is the bias term for the output values ​​of different indicators in the energy conversion group corresponding to the v-th neuron in the second hidden layer;

[0147] When s3=10,12, v∈{1,2,…,50}; when s3=11, v∈{36,37,…,50}.

[0148] The energy conversion group includes three indicators: thermal efficiency, combustion efficiency, and heat loss rate, denoted as Y10~Y12 respectively. These indicators characterize energy conversion from different perspectives: thermal efficiency reflects the overall energy utilization level, combustion efficiency reflects the integrity of chemical energy conversion, and heat loss rate characterizes process energy loss. These three indicators are directly related to the input calorific value and temperature parameters, and can comprehensively evaluate the energy utilization effect of the heat treatment process. Thermal efficiency Y10 and heat loss rate Y12 need to consider the energy conversion of the entire heat treatment process, and therefore are related to all process groups; while combustion efficiency Y11 specifically reflects the energy conversion effect of the combustion stage, and therefore is only related to the combustion process group.

[0149] S5. Design the loss function for a deep learning-based prediction model of solid waste thermal treatment;

[0150] Furthermore, the loss function in step S5 considers both prediction loss and heat treatment loss. The specific implementation method includes the following steps:

[0151] S5.1. Construct the prediction loss based on the conversion efficiency prediction loss, product distribution prediction loss, and energy conversion prediction loss, with the following expression:

[0152]

[0153] in, Predicting losses for conversion efficiency; Predicting losses for product distribution; Energy conversion prediction loss;

[0154]

[0155] in, This is the function for calculating the mean square error; Output values ​​for different indicators in the conversion efficiency group predicted by the model; Output values ​​for different indicators in the actual measured conversion efficiency group;

[0156]

[0157] in, Output values ​​for different indicators in the product distribution group predicted by the model; Output values ​​for different indicators in the actual measured product distribution group;

[0158]

[0159] in, Output values ​​for different indicators in the energy conversion group predicted by the model; These are the output values ​​of different indicators in the actual measured energy conversion group;

[0160] S5.2. Constructing heat treatment loss based on material conversion constraints, composition conversion constraints, and heat conversion constraints. The expression is:

[0161]

[0162] in, Material conversion constraints The weights corresponding to material conversion constraints; Constraints on component transformation The weights corresponding to the component transformation constraints; For heat conversion constraints, The weights corresponding to the heat conversion constraints;

[0163] Material conversion constraints are established as follows:

[0164]

[0165] in, For volatile matter conversion terms; For fixed carbon allocation items;

[0166] Furthermore, in the material conversion constraints, two key material conversion balance relationships are mainly considered: one is the conversion balance of volatiles, that is, the material balance of volatiles in the raw material after being converted into gaseous products through pyrolysis and entering the flue gas, which is characterized by comparing the difference between the amount of volatiles converted in the feed and the amount of volatiles finally entering the flue gas; the other is the distribution balance of fixed carbon, that is, the distribution balance of fixed carbon in the raw material in the slag and fly ash after combustion, which is characterized by comparing the difference between the content of fixed carbon in the feed and the amount of fixed carbon finally distributed in the slag and fly ash.

[0167] The component transformation constraint is established as follows:

[0168]

[0169] in, Constraints on sulfur element transformation Constraints for chlorine element transformation;

[0170] The heat conversion constraint is established as follows:

[0171] ;

[0172] In the component transformation constraint, the transformation law of two key elements in the high-temperature heat treatment process is mainly considered: the material balance of sulfur element being converted into SO2 at high temperature and entering the flue gas is characterized by comparing the difference between the amount of sulfur element converted in the feed and the amount of sulfur element entering the flue gas after SO2 is finally formed; the material balance of chlorine element being converted into HCl during pyrolysis and combustion and entering the flue gas is characterized by comparing the difference between the amount of chlorine element converted in the feed and the amount of chlorine element entering the flue gas after HCl is finally formed.

[0173] S5.3. Construct the total loss based on predicted loss and heat treatment loss. ;

[0174]

[0175] in, The weights corresponding to the predicted loss, The weights corresponding to heat treatment losses are determined by expert experience.

[0176] S6. Connect the input layer, first hidden layer, second hidden layer and output layer of the deep learning-based solid waste thermal treatment prediction model in sequence to obtain the deep learning-based solid waste thermal treatment prediction model.

[0177] S7. Collect data through the solid waste treatment plant and automated control system, normalize the data, divide the data into training set, validation set and test set in a ratio of 7:2:1, train, validate and test the deep learning-based solid waste thermal treatment prediction model obtained in step S6, and obtain the trained deep learning-based solid waste thermal treatment prediction model.

[0178] Furthermore, the parameter settings need to include hyperparameters that need to be manually set based on expert experience, such as network structure parameters and the number of neurons in each layer; hyperparameters that need to be manually set during model training, such as learning rate, batch size, training epochs, and weight coefficients in the loss function; and parameters that need to be determined by the model during training, such as weight coefficients and bias terms in the model.

[0179] Furthermore, in step S7, the model training uses the backpropagation algorithm. During training, the model calculates the predicted value through forward propagation, substitutes the predicted value and the true value into the loss function to calculate the loss, and then updates the network parameters through backpropagation. To prevent overfitting, an early stopping strategy is adopted, and training is stopped when the validation set loss does not decrease for 10 consecutive epochs. The output result is restored to the actual physical quantity through inverse normalization.

[0180] S8. After inverse normalization of the prediction results of the deep learning-based solid waste thermal treatment prediction model, the prediction results of the deep learning-based solid waste thermal treatment prediction model with actual physical quantities are obtained. An anomaly prediction method for solid waste thermal treatment is constructed. When considering the anomaly problem of solid waste thermal treatment, the analysis is carried out from the perspectives of abnormal material transformation, abnormal material distribution, and abnormal energy transformation, and an anomaly diagnosis and judgment scheme is given.

[0181] In the thermal treatment of solid waste, due to the complex composition of solid waste and the large fluctuation of its calorific value, the thermal treatment process includes multiple stages such as drying, pyrolysis, and combustion, and there are strong coupling relationships between these stages. Therefore, in order to ensure the normal operation of solid waste thermal treatment, higher requirements are needed for operation management, and it is necessary to promptly detect abnormal problems in the solid waste thermal treatment process and guide optimization and adjustment decisions through prediction.

[0182] Furthermore, the specific implementation method of step S8 includes the following steps:

[0183] S8.1. In the process of identifying anomalies in the thermal treatment of solid waste, the degree of material transformation anomaly reflects the degree of independent deviation of each process parameter, which helps to accurately locate the abnormal link and take targeted adjustments. The calculation formula is:

[0184]

[0185] in, Output values ​​for different indicators predicted by the model. include , , ; To and The corresponding standard values ​​are determined by expert experience, design documents, and experimental research methods; The allowable fluctuation range corresponding to the formula for the degree of material transformation anomaly is determined by expert experience, design documents, experimental research, and other methods. for and The corresponding allocation weight coefficients;

[0186] S8.2. The anomaly degree of material distribution includes the anomaly degree of flue gas distribution and the anomaly degree of solid distribution;

[0187] Considering the distribution deviations of each component in the flue gas during moisture gasification, volatile matter desorption, carbon oxidation, sulfur oxidation, and hydrogen chloride generation, calculate the flue gas distribution anomaly degree. The expression is:

[0188]

[0189] in, To assign weighting coefficients, s2 = 71, 72, 73, 74, 75; To and The corresponding standard values ​​are determined by expert experience, design documents, experimental research, and other methods. This represents the allowable fluctuation range corresponding to the flue gas distribution anomaly formula.

[0190] Considering the distribution equilibrium of inorganic matter in slag and fly ash, and combining the influence of process parameters such as combustion temperature, residence time, and gas-solid separation efficiency on the slag-to-fly ash ratio, the degree of solid distribution anomaly is calculated. The expression is:

[0191]

[0192] in, , The weighting coefficients are assigned to slag and fly ash, respectively. , These represent the allowable fluctuation ranges corresponding to the weighting coefficients of slag and fly ash in the solid distribution anomaly formula;

[0193] S8.3. Energy conversion anomaly degree includes energy anomaly degree. and energy balance anomaly ;

[0194] Considering the impact of factors such as solid waste calorific value fluctuations, combustion air distribution, furnace temperature distribution, and heat exchange efficiency on system thermal efficiency, combustion efficiency, and heat loss, the energy anomaly degree is calculated. The expression is:

[0195]

[0196] in, For the energy index weights, s3 = 10, 11, 12; This represents the allowable fluctuation range corresponding to the energy anomaly formula;

[0197] Considering the balance between effective heat output and heat loss in the system, and taking into account factors such as furnace wall insulation performance and cooling system efficiency, the energy balance anomaly degree is calculated. The expression is:

[0198]

[0199] in, The allowable error for energy balance is determined by expert experience, design documents, experimental research, and other methods.

[0200] S8.4. Construct the comprehensive anomaly index ZH;

[0201]

[0202] in, , , , , They are respectively with , , , , The corresponding process weighting coefficients are determined by expert experience, design documents, experimental research, and other methods.

[0203] S8.5. Based on the comprehensive abnormality index obtained in step S4, perform anomaly diagnosis and judgment;

[0204] The anomaly levels are divided into four categories: normal operation, minor anomaly, moderate anomaly, and severe anomaly. The critical comprehensive anomaly indices corresponding to minor, moderate, and severe anomalies are as follows: , , Then we get For normal operation; It is a minor abnormality; Moderately abnormal; This is a serious abnormality.

[0205] This embodiment proposes a deep learning model construction method for analyzing the characteristics of solid waste thermal treatment process. The method has made targeted designs and optimizations for the thermal treatment characteristics of solid waste in the input parameters, the first hidden layer, the second hidden layer, and the output layer, realizing the scientificity and rationality of the prediction of conversion efficiency, product distribution and energy conversion in the thermal treatment process of solid waste.

[0206] This embodiment proposes a method for predicting anomalies in the thermal treatment of solid waste. Utilizing an established deep learning model, the method determines the state of solid waste thermal treatment from the perspectives of abnormal material transformation, abnormal material distribution, and abnormal energy conversion. This allows for a comprehensive evaluation of system performance, timely detection and handling of potential operational problems, and ensures the continuity and safety of the solid waste treatment process.

[0207] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0208] Although this application has been described above with reference to specific embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of this application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in this application can be combined with each other in any way. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, this application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning, characterized in that, Includes the following steps: S1. Determine the input parameters of the input layer of the deep learning-based solid waste thermal treatment prediction model, including material characteristic parameters, process operation parameters, and process parameters; S2. Construct the first hidden layer of a deep learning-based prediction model for the thermal treatment of solid waste; Step S2 divides the nodes of the first hidden layer into a material characteristic group, a reaction characteristic group, and an energy characteristic group. The material characteristic group focuses on the changes in the physicochemical properties of materials during heat treatment and contains 30 neuron nodes, with the i1th neuron node having a value of 1 to 30. The reaction characteristic group focuses on describing the extent of various chemical reactions and contains 25 neuron nodes, with the i2th neuron node having a value of 31 to 55. The energy characteristic group focuses on characterizing energy conversion and transfer features and contains 25 neuron nodes, with the i3th neuron node having a value of 56 to 80. S2.

1. Temperature correction and material moisture correction are introduced into the material property group. Temperature correction considers the driving effect of temperature on the material conversion rate and reflects the differences in the material conversion law in different temperature ranges; material moisture correction considers the influence of the moisture evaporation process on subsequent conversion. S2.

2. Constructing a reaction characteristic group introduces a mass conversion term to directly reflect the degree of reaction of substances and embody the depth of the reaction; adding an excess air term characterizes the sufficiency of the combustion environment and reflects the influence of oxidation reaction conditions on the conversion effect; S2.

3. Construct an energy characteristic group by introducing a calorific value conversion term to directly characterize the energy conversion effect; add a temperature gradient term to reflect the driving force and intensity of heat transfer, and describe the spatial distribution pattern of energy; S3. Construct the second hidden layer of a deep learning-based prediction model for the thermal treatment of solid waste; Step S3 divides the nodes of the second hidden layer into a drying process group, a pyrolysis process group, and a combustion process group. The drying process group describes the influence of the drying process on subsequent pyrolysis and combustion, and contains 15 neuron nodes, with the i4th neuron node having a value of 1 to 15. The pyrolysis process group describes the compositional changes of the pyrolysis products, and contains 20 neuron nodes, with the i5th neuron node having a value of 16 to 35. The combustion process group describes the process characteristics of the final oxidation stage, and contains 15 neuron nodes, with the i6th neuron node having a value of 36 to 50. S3.

1. Constructing a drying process group: By characterizing the process features of moisture removal from solid waste, reflecting the driving effect of temperature on moisture evaporation, and describing the impact of the drying process on subsequent pyrolysis and combustion; S3.

2. Constructing a pyrolysis process group: By reflecting the kinetic characteristics of organic matter pyrolysis and decomposition, the influence of temperature on volatile matter is analyzed, the compositional changes of pyrolysis products are described, and key characteristics of the pyrolysis stage are extracted. S3.

3. Construct a combustion process group to characterize the oxidation process of fixed carbon and volatiles, reflect the effect of oxygen concentration on combustion efficiency, and describe the process characteristics of the final oxidation stage; S4. Construct the output layer of the deep learning-based solid waste thermal treatment prediction model, divide the output layer into conversion efficiency group, product distribution group and energy conversion group, and determine the output parameters of the output layer of the deep learning-based solid waste thermal treatment prediction model. S5. Design the loss function for a deep learning-based prediction model of solid waste thermal treatment; S6. Connect the input layer, first hidden layer, second hidden layer and output layer of the deep learning-based solid waste thermal treatment prediction model in sequence to obtain the deep learning-based solid waste thermal treatment prediction model. S7. Collect data through the solid waste treatment plant and automated control system, normalize the data, divide the data into training set, validation set and test set in a ratio of 7:2:1, train, validate and test the deep learning-based solid waste thermal treatment prediction model obtained in step S6, and obtain the trained deep learning-based solid waste thermal treatment prediction model. S8. After inverse normalization of the prediction results of the deep learning-based solid waste thermal treatment prediction model, the prediction results of the deep learning-based solid waste thermal treatment prediction model with actual physical quantities are obtained. An anomaly prediction method for solid waste thermal treatment is constructed. When considering the anomaly problem of solid waste thermal treatment, the analysis is carried out from the perspectives of abnormal material transformation, abnormal material distribution, and abnormal energy transformation, and an anomaly diagnosis and judgment scheme is given.

2. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 1, characterized in that, In step S1, the number of nodes in the input layer is determined to be 35, denoted as X1-X2. 35 ; X1~X 14 These are material characteristic parameters, with the specific parameters as follows: X1 is moisture content, X2 is ash content, X3 is combustible content, X4 is carbon content, X5 is hydrogen content, X6 is oxygen content, X7 is nitrogen content, X8 is sulfur content, X9 is chlorine content, X... 10 X represents the calorific value. 11 X is the bulk density. 12 For particle size, X 13 X represents the sediment content. 14 Metal content; X 15 ~X 25 These are the process operation parameters, and the specific parameters are as follows: X 15 X is the grate speed. 16 X is the feed rate. 17 For primary air volume, X 18 For secondary air volume, X 19 X represents the furnace temperature. 20 For negative pressure in the furnace, X 21 X represents the flue temperature. 22 X represents the steam temperature. 23 X is the steam pressure. 24 X represents the steam flow rate. 25 The ambient temperature; X 26 ~X 35 These are process parameters, and the specific parameters are as follows: X 26 X represents the feed moisture content. 27 X represents the discharge moisture content. 28 For feed volatiles, X 29 For the discharge volatile matter, X 30 X is the feed mass flow rate. 31 X represents the discharge mass flow rate. 32 X is the calorific value of the feed. 33 X is the discharge calorific value. 34 X represents the oxygen concentration in the smoke. 35 This refers to the ash content.

3. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 2, characterized in that, The specific implementation method of step S2 includes the following steps: S2.

1. The calculation method for the 30 neuron nodes in the material property group is as follows: ； in, This represents the output value of the i1th neuron node in the first hidden layer, where i1 ranges from 1 to 30. This is an activation function used to introduce nonlinear features; The weight coefficients are the weights of the i1th neuron node in the first hidden layer corresponding to the jth input parameter. For input parameters, ; This is the bias term for the i1th neuron node in the first hidden layer corresponding to the jth input parameter; The temperature correction factor is determined by expert experience, design documents, and experimental research methods. The moisture correction factor is determined by expert experience, design documents, and experimental research methods. This is a temperature correction term; This is a correction item for material moisture content; S2.

2. The calculation method for the 25 neuron nodes in the response characteristic group is as follows: ； in, This is the output value of the i2th neuron node in the first hidden layer, where i2 takes values ​​from 31 to 55; The weight coefficients are the weights of the i2-th neuron node in the first hidden layer corresponding to the j-th input parameter. This is the bias term for the i2-th neuron node in the first hidden layer corresponding to the j-th input parameter; The quality conversion coefficient is determined by expert experience, design documents, and experimental research methods. The excess air coefficient is determined by expert experience, design documents, and experimental research methods. The density of air is determined by actual measurement. For quality conversion items; This is an item indicating excess air. S2.

3. The calculation method for the 25 neuron nodes in the energy characteristic group is as follows: ； in, This is the output value of the i3th neuron node in the first hidden layer, where i3 takes values ​​from 56 to 80; The weight coefficients of the j-th input parameter and the corresponding i3-th neuron node in the first hidden layer; This is the bias term for the i3rd neuron node in the first hidden layer corresponding to the jth input parameter; This is a calorific value conversion term; The calorific value conversion coefficient is determined by expert experience, design documents, and experimental research methods. This is the temperature gradient term; The temperature gradient coefficient is determined by expert experience, design documents, and experimental research methods.

4. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 3, characterized in that, The specific implementation method of step S3 includes the following steps: S3.

1. The calculation method for the 15 neuron nodes in the drying process group is as follows: ； in, This is the output value of the i4th neuron node in the second hidden layer, where i4 takes a value from 1 to 15; These are the weight coefficients of the i4th neuron node in the second hidden layer, corresponding to the kth neuron in the first hidden layer. =1~80; This is the bias term for the i4th neuron node in the second hidden layer, which corresponds to the kth neuron in the first hidden layer; This represents the output value of the k-th neuron in the first hidden layer. This is the drying process item. The drying process coefficient is determined by expert experience, design documents, and experimental research methods. S3.

2. The calculation method for the 20 neuron nodes in the pyrolysis process group is as follows: ； in, This is the output value of the i5th neuron node in the second hidden layer, where i5 ranges from 16 to 35. The weight coefficients are the weights of the i5th neuron node in the second hidden layer, corresponding to the kth neuron in the first hidden layer. This is the bias term for the i5th neuron node in the second hidden layer, which corresponds to the kth neuron in the first hidden layer; This is the pyrolysis process term. The coefficient for the pyrolysis process is determined by expert experience, design documents, and experimental research methods. S3.

3. The calculation method for the 15 neuron nodes in the combustion process group is as follows: ； in, This is the output value of the i6th neuron node in the second hidden layer, where i6 takes a value of 36 to 50; The weight coefficients are the weights of the i6th neuron node in the second hidden layer, corresponding to the kth neuron in the first hidden layer. This is the bias term for the i6th neuron node in the second hidden layer, which corresponds to the kth neuron in the first hidden layer; For the combustion process, The combustion process coefficient, The reference oxygen concentration is determined by expert experience, design documents, and experimental research methods.

5. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 4, characterized in that, Step S4, based on the evaluation requirements of the solid waste thermal treatment process, divides the output layer into a conversion efficiency group, a product distribution group, and an energy conversion group. The conversion efficiency group index reflects the completeness and sufficiency of material conversion, the product distribution group index reveals the migration law of materials and environmental impact, and the energy conversion group index reflects the energy utilization level. The conversion efficiency group indicators include Y1~Y6, where Y1 is the moisture conversion rate, Y2 is the volatile matter conversion rate, Y3 is the fixed carbon conversion rate, Y4 is the carbon conversion rate, Y5 is the sulfur conversion rate, and Y6 is the chlorine conversion rate. Product distribution group indicators include the material distribution ratio indicator Y. 71 ~ Y 75 And residue allocation index Y8~Y9, Y 71 Y represents the mass distribution ratio of moisture entering the flue gas after drying. 72 Y represents the mass distribution ratio of volatile matter entering the flue gas after pyrolysis. 73 Y represents the proportion of carbon elements converted into carbon dioxide that enters the flue gas. 74 Y represents the proportion of sulfur that has been converted into sulfur dioxide and enters the flue gas. 75 Y8 represents the mass distribution ratio of chlorine after it is converted into hydrogen chloride and enters the flue gas; Y9 represents the mass distribution ratio of fixed carbon residue in the slag after combustion; and Y9 represents the mass distribution ratio of fly ash components after the combustion of fixed carbon and enters the fly ash. Energy conversion group indicators include Y 10 ~Y 12 Y 10 For thermal efficiency, Y 11 For combustion efficiency, Y 12 This refers to the heat loss rate; The specific implementation method of step S4 includes the following steps: S4.

1. The calculation method for the output values ​​of different indicators in the conversion efficiency group is as follows: ； in, s1 represents the output values ​​of different indicators in the conversion efficiency group; , where is the weighting coefficient of the output values ​​of different indicators in the conversion efficiency group corresponding to the v-th neuron in the second hidden layer; This is a bias term for the output values ​​of different indicators in the conversion efficiency group corresponding to the v-th neuron in the second hidden layer; This represents the output value of the v-th neuron in the second hidden layer. When s1=1, v∈{1,2,…,15}; when s1=2, v∈{16,17,…,35}; when s1=3~6, v∈{36,37,…,50}. S4.

2. The calculation method for the output values ​​of different indicators in the product distribution group is as follows: ； in, s1 represents the output values ​​of different indicators in the product distribution group, and s2 represents different indicators in the product distribution group. , where is the weighting coefficient of different index output values ​​in the product distribution group corresponding to the v-th neuron in the second hidden layer; This is the bias term for the output values ​​of different indicators in the product distribution group corresponding to the v-th neuron in the second hidden layer; When s2=71, v∈{1,2,…,15}; when s2=72, v∈{16,17,…,35}; when s2=73,74,75,8,9, v∈{36,37,…,50}. S4.

3. The calculation method for the output values ​​of different indicators in the energy conversion group is as follows: ； in, s1 represents the output value of different indicators in the energy conversion group, and s2 represents different indicators in the energy conversion group. , where is the weighting coefficient of different index output values ​​in the energy conversion group corresponding to the v-th neuron in the second hidden layer; This is the bias term for the output values ​​of different indicators in the energy conversion group corresponding to the v-th neuron in the second hidden layer; When s3=10,12, v∈{1,2,…,50}; when s3=11, v∈{36,37,…,50}.

6. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 5, characterized in that, The loss function in step S5 considers both prediction loss and heat treatment loss. The specific implementation method includes the following steps: S5.

1. Construct the prediction loss based on the conversion efficiency prediction loss, product distribution prediction loss, and energy conversion prediction loss, with the following expression: ； in, Predicting losses for conversion efficiency; Predicting losses for product distribution; Energy conversion prediction loss; ； in, This is the function for calculating the mean square error; Output values ​​for different indicators in the conversion efficiency group predicted by the model; Output values ​​for different indicators in the actual measured conversion efficiency group; ； in, Output values ​​for different indicators in the product distribution group predicted by the model; Output values ​​for different indicators in the actual measured product distribution group; ； in, Output values ​​for different indicators in the energy conversion group predicted by the model; These are the output values ​​of different indicators in the actual measured energy conversion group; S5.

2. Constructing heat treatment loss based on material conversion constraints, composition conversion constraints, and heat conversion constraints. The expression is: ； in, Material conversion constraints The weights corresponding to material conversion constraints; Constraints on component transformation The weights corresponding to the component transformation constraints; For heat conversion constraints, The weights corresponding to the heat conversion constraints; Material conversion constraints are established as follows: ； in, For volatile matter conversion terms; For fixed carbon allocation items; The component transformation constraint is established as follows: ； in, Constraints on sulfur element transformation Constraints for chlorine element transformation; The heat conversion constraint is established as follows: ； S5.

3. Construct the total loss based on predicted loss and heat treatment loss. ; ； in, The weights corresponding to the predicted loss, The weights corresponding to heat treatment losses are determined by expert experience.

7. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 6, characterized in that, In step S7, the model training uses the backpropagation algorithm. During training, the model calculates the predicted value through forward propagation, substitutes the predicted value and the true value into the loss function to calculate the loss, and then updates the network parameters through backpropagation. To prevent overfitting, an early stopping strategy is adopted, and training is stopped when the validation set loss does not decrease for 10 consecutive epochs. The output result is restored to the actual physical quantity through inverse normalization.

8. The method for predicting and identifying anomalies in the thermal treatment of solid waste based on deep learning according to claim 7, characterized in that, The specific implementation method of step S8 includes the following steps: S8.

1. In the process of identifying anomalies in the thermal treatment of solid waste, the degree of material transformation anomaly reflects the degree of independent deviation of each process parameter, which helps to accurately locate the abnormal link and take targeted adjustments. The calculation formula is: ； in, Output values ​​for different indicators predicted by the model. include , , ; To and The corresponding standard values ​​are determined by expert experience, design documents, and experimental research methods; The allowable fluctuation range corresponding to the formula for the degree of anomaly in material transformation is determined by expert experience, design documents, and experimental research methods. for and The corresponding allocation weight coefficients; S8.

2. The anomaly degree of material distribution includes the anomaly degree of flue gas distribution and the anomaly degree of solid distribution; Considering the distribution deviations of each component in the flue gas during moisture gasification, volatile matter desorption, carbon oxidation, sulfur oxidation, and hydrogen chloride generation, calculate the flue gas distribution anomaly degree. The expression is: ； in, To assign weighting coefficients, s2 = 71, 72, 73, 74, 75; To and The corresponding standard values ​​are determined by expert experience, design documents, and experimental research methods; This represents the allowable fluctuation range corresponding to the flue gas allocation anomaly formula. Considering the distribution equilibrium of inorganic matter in slag and fly ash, and combining the effects of combustion temperature, residence time, and gas-solid separation efficiency on the slag-to-fly ash ratio, the degree of solid distribution anomaly is calculated. The expression is: ； in, , The weighting coefficients are assigned to slag and fly ash, respectively. , These represent the allowable fluctuation ranges corresponding to the weighting coefficients of slag and fly ash in the solid distribution anomaly formula; S8.

3. Energy conversion anomaly degree includes energy anomaly degree. and energy balance anomaly ; Considering the impact of solid waste calorific value fluctuations, combustion air distribution, furnace temperature distribution, and heat exchange efficiency on system thermal efficiency, combustion efficiency, and heat loss, calculate the energy anomaly degree. The expression is: ； in, For the energy index weights, s3 = 10, 11, 12; This represents the allowable fluctuation range corresponding to the energy anomaly formula; Considering the balance between effective heat output and heat loss in the system, and taking into account the effects of furnace wall insulation performance and cooling system efficiency, the energy balance anomaly degree is calculated. The expression is: ； in, The allowable error for energy balance is determined by expert experience, design documents, and experimental research methods. S8.

4. Construct the comprehensive anomaly index ZH; ； in, , , , , They are respectively with , , , , The corresponding process weighting coefficients are determined by expert experience, design documents, and experimental research methods; S8.

5. Based on the comprehensive abnormality index obtained in step S4, perform anomaly diagnosis and judgment; The anomaly levels are divided into four categories: normal operation, minor anomaly, moderate anomaly, and severe anomaly. The critical comprehensive anomaly indices corresponding to minor, moderate, and severe anomalies are as follows: , , Then we get For normal operation; It is a minor abnormality; Moderately abnormal; This is a serious abnormality.

Images