Energy-saving treatment method and device for high-water-content sludge based on smoldering and ai prediction

Abstract

This invention discloses an energy-saving treatment method and apparatus for high-moisture sludge based on smoldering and AI prediction. Addressing the high energy consumption problem of traditional incineration of 80% high-moisture sludge, this invention reveals the dual-peak competition of DTG and the regulation of moisture content. When the dual-path exothermic reaction of heterogeneous coke oxidation and homogeneous pyrolysis gas oxidation overcomes heat dissipation and evaporation heat, a self-sustaining pyrolysis-oxidation coupled cycle is established, with a burnout rate greater than 95%. The method includes acquiring TG / DTG curves and flue gas and temperature field data; extracting and optimizing kinetic parameters based on the three-step smoldering reaction; constructing a multi-dimensional database; employing the XGBoost algorithm with the loss function and regularization term as objective functions; accelerating the solution with second-order Taylor expansion and suppressing overfitting; achieving a prediction error of less than 5%; establishing a self-sustaining critical spectrum; and iteratively optimizing to obtain the globally optimal operating condition. This invention achieves low-energy clean treatment and intelligent selection of process parameters through the collaborative development of a smoldering self-sustaining treatment device and an AI prediction model.

Description

Technical Field

[0001] This invention relates to the field of sludge treatment technology, and in particular to an energy-saving method and apparatus for treating high-moisture sludge based on smoldering and AI prediction. Background Technology

[0002] Traditional thermal treatment of sludge primarily involves incineration, which reduces sludge volume and renders it harmless through flame combustion. However, Hao et al. (2020) pointed out that the energy consumption of direct incineration of dewatered sludge exceeds the recoverable energy, making it a net energy-consuming process. Self-sustaining combustion requires the sludge moisture content to be reduced to below 50%. Municipal sludge, even after mechanical dewatering, typically still has a moisture content of around 80%, far from meeting the conditions for self-sustaining combustion. Therefore, pre-drying or co-firing with auxiliary fuels is necessary, resulting in high system energy consumption and poor economic efficiency. Ci Hanlin et al. (2021) further analyzed the carbon footprint of the entire sludge drying-incineration chain, showing that the carbon emissions from the dewatering and drying stage account for more than 60% of the total carbon emissions, making it the core contributor to energy consumption and carbon footprint in the entire process. Furthermore, the incineration process involves high temperatures (>800℃), where N2 and O2 in the air react to generate NO at high temperatures. x The flue gas purification system is complex. Smoldering, as a slow, low-temperature, flameless combustion method, has a minimum ignition heat flux of only about 25±5 kW / m³. 2 The moisture content of smoldering is significantly lower than that of flaming combustion, and once self-sustaining propagation is established, the exothermic oxidation reaction can drive the pyrolysis reaction to continue, forming a positive feedback loop of pyrolysis-oxidation coupling, without the need for continuous external energy input. In summary, while traditional flaming combustion routes, represented by incineration, can achieve solid waste reduction and harmlessness, they have fundamental limitations when dealing with solid waste with high moisture content—self-sustaining combustion cannot be maintained, pre-drying energy consumption is too high, and high-temperature conditions inevitably lead to secondary NOx generation. All three problems stem from the strict limitations of flaming combustion on moisture content and its rigid dependence on high temperatures. Smoldering theoretically overcomes these bottlenecks, but the quantitative mapping relationship between moisture content and reaction rate has not yet been established, and the accurate characterization of self-sustaining criteria and the determination of optimal aeration strategies under different moisture contents still face challenges.

[0003] Currently, research on smoldering both domestically and internationally mainly focuses on the ignition limits and combustion behavior of single combustibles such as forest peat, wood, and asphalt. Research on the kinetic mechanisms of smoldering treatment of sludge is still in its early stages. Existing studies mostly use cone calorimeters and thermogravimetric analysis experiments to obtain macroscopic data such as mass loss, temperature changes, and gas emissions, supplemented by numerical simulation methods to fit reaction kinetic parameters. Zhang et al. conducted experimental and modeling analyses on the smoldering behavior of municipal dewatered sludge under different gas flow reaction rates, and derived analytical expressions for smoldering temperature and propagation reaction rate using the activation energy asymptotic method. However, this method can only describe the average smoldering characteristics under a single variable and cannot distinguish the individual contributions of multi-step coupled processes such as pyrolysis and oxidation. Zanoni et al. constructed a two-dimensional numerical model of the smoldering system, coupling heat and mass transfer with one-step oxidation reaction kinetics, revealing the significant impact of radial heat loss on smoldering propagation. However, such CFD numerical simulation methods have inherent limitations: fully coupled two-dimensional transient simulations require sub-millimeter grids to cover hours of physical time, and a single calculation for a given operating condition can take tens of hours, making it almost infeasible for traversal scans of multivariate combined operating conditions; furthermore, the model only outputs results for a single operating condition each time, lacking the ability to quickly extrapolate the effects of treatments for unexperimented conditions based on existing data. Therefore, current research lacks the integration of experimental data on smoldering reaction kinetics with AI prediction methods. In complex systems with multiple interactive influences, the reliance on trial-and-error experimental-simulation methods makes it difficult to quickly select optimal process parameters, thus hindering the large-scale engineering application of smoldering sludge treatment technology.

[0004] In summary, existing solid waste thermal treatment technologies suffer from poor adaptability to high moisture content and high energy consumption. While smoldering possesses the theoretical advantage of low-energy treatment of high moisture content solid waste, its kinetic mechanism is not yet fully understood, particularly lacking intelligent modeling methods that combine AI prediction techniques with smoldering experimental data. This hinders the rapid selection of optimal process parameters under various operating conditions. Therefore, there is an urgent need to establish a treatment method for high moisture content organic solid waste based on smoldering reaction kinetics and AI prediction to overcome current technological bottlenecks. Summary of the Invention

[0005] The purpose of this invention is to provide an energy-saving method and apparatus for treating high-moisture sludge based on smoldering and AI prediction, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides an energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction, comprising the following steps: S1. Conduct smoldering and heating experiments under different working conditions on sludge samples with different moisture contents, clarify the critical conditions for smoldering self-sustaining under different moisture contents, and obtain thermogravimetric curves and differential thermogravimetric curves in heating experiments under different working conditions. S2. Measure the changes in flue gas composition, temperature field, and mass during the smoldering experiment to obtain data on flue gas composition, temperature field, and mass changes; S3. Based on the three-step smoldering reaction principle, and using the obtained thermogravimetric curves and differential thermogravimetric curves, calculate the kinetic parameters of the smoldering reaction, and optimize the kinetic parameters using the least squares objective function. S4. Organize and preprocess the experimental data obtained from steps S1 to S3, construct a multidimensional database, and divide the multidimensional database into a training set, a validation set, and a test set. S5. Construct a prediction model. Train the prediction model using the training set, optimize by minimizing the least squares objective function, and fine-tune the hyperparameters of the trained prediction model using the validation set to obtain a high-generalization prediction model. S6. The generalization ability of the high generalization prediction model is evaluated using a test set. Based on the evaluated prediction model, predictions are made for different combinations of working conditions for sludge with high water content. A smoldering self-sustaining criticality prediction spectrum is established based on the prediction results. S7. Based on the established smoldering self-sustaining criticality prediction map, the prediction model is verified and optimized. Based on the optimized prediction model, the global operating space of high water content sludge is predicted to obtain the optimal operating condition map.

[0007] Preferably, the operating conditions include radiation intensity and heating rate.

[0008] Preferably, step S3 specifically includes: S31. Define the conversion rate based on thermogravimetric curve data, normalize the reaction process, and determine the reaction state; S32. Based on the three-step smoldering reaction principle, estimate the kinetic parameters of pyrolysis, sludge oxidation and carbon oxidation reactions based on Arrhenius law and the peak value of differential thermogravimetric curves. The three-step smoldering reaction includes water evaporation, pyrolysis and oxidation, and oxidation includes sludge oxidation and carbon oxidation. S33. The kinetic parameters of the pyrolysis, sludge oxidation and carbon oxidation reactions are optimized and solved using the least squares objective function to obtain the kinetic parameters of each reaction.

[0009] Preferably, the kinetic parameters include the pre-exponential factor and the activation energy.

[0010] The preferred least squares objective function is: ; In the formula, The parameter vector to be optimized includes the pre-exponential factors, activation energies, and stoichiometric coefficients for pyrolysis, sludge oxidation, and carbon oxidation reactions. N For the number of experimental data points, These are experimental values. For the corresponding simulated values, This is the error normalization scaling factor.

[0011] Preferably, the prediction model is constructed using the XGBoost extreme gradient boosting algorithm. The input feature vector includes thermogravimetric curve data, differential thermogravimetric curve data, heating rate, and radiation intensity under different moisture content conditions. The thermogravimetric curve data includes mass loss, and the differential thermogravimetric curve data includes mass loss reaction rate. The output target vector includes mass loss and mass loss reaction rate. The loss function consists of a loss term and a regularization term.

[0012] Preferably, the evaluation indicators include the coefficient of determination, root mean square error, and point-by-point deviation between the thermogravimetric and differential thermogravimetric experimental curves and the fitted curves of the prediction model.

[0013] Preferably, the smoldering self-sustaining criticality prediction spectrum is established based on radiation intensity-ignition time-moisture content, wherein there is a double logarithmic linear relationship between radiation intensity and ignition time.

[0014] Preferably, step S7 specifically includes: A new operating condition not included in either the training or test set was selected for supplementary verification experiments; The thermogravimetric curves and differential thermogravimetric curves under the new operating conditions are compared point by point with the fitted curves of the predicted values ​​of the prediction model, and the average deviation over the entire temperature range is calculated. If the average deviation between the prediction and the experiment exceeds the preset accuracy threshold, the experimental data for that working condition is added to the training database, the prediction model hyperparameters are iteratively optimized, and the prediction model is retrained until the prediction deviation of the prediction model for the new working condition meets the accuracy requirements, at which point the iteration stops.

[0015] Preferably, the deviation is defined as: ; In the formula, ε Prediction bias is expressed as a percentage. M To verify the total number of experimental data points, For the first The model prediction value for each data point. This is a supplementary experimental measurement for the j-th data point.

[0016] This invention also provides an energy-saving treatment device for high-moisture sludge based on smoldering and AI prediction, including a computer, a data acquisition unit, a control unit, a control valve, a gas composition analyzer, a temperature sensor, a heat exchange tube, an igniter, an airflow mixer, a vent, a flow meter, additional fuel, a gas flow control unit, an air tank, and a heating wire. The data acquisition unit is connected to the temperature sensor, which is embedded inside the sludge and ash. The data acquisition unit is connected to the computer, and the computer is connected to the control unit. The control unit adjusts the opening of the air tank and additional fuel pipelines by regulating the control valve. The flow meter is installed on the air tank and additional fuel pipelines, both of which are connected to the airflow mixer. The igniter is located above the sludge, and the airflow mixer is connected to the igniter. The vent is located at the bottom of the ash, which is located below the sludge. The heating wire is located below the vent. The gas flow control unit is connected to the flow meter to control the supply of additional fuel. The heat exchange tube absorbs the waste heat from the high-temperature flue gas at its heat-absorbing end and transfers heat back to the ash at its heat-releasing end. The gas composition analyzer analyzes the composition of the gas after combustion.

[0017] Therefore, the energy-saving treatment method and apparatus for high-moisture sludge based on smoldering and AI prediction, as described above, have the following beneficial effects: (1) The XGBoost extreme gradient boosting tree intelligent prediction model is adopted. The thermogravimetric curve data and differential thermogravimetric curve data of sludge with different moisture contents are used as input features to perform regression prediction on the smoldering reaction kinetic parameters and treatment effect indicators. The smoldering self-sustaining critical prediction spectrum of radiation intensity-ignition time-moisture content is established to realize the rapid determination of ignition time and intelligent screening of optimal parameters under different working conditions. (2) Based on the smoldering pyrolysis-oxidation coupling kinetics, this invention requires only a very low heat flux to trigger the self-sustaining propagation of smoldering, with a burnout rate of >95%. Compared with traditional direct incineration, this method significantly reduces energy consumption, simplifies the operation process, greatly reduces the application difficulty of smoldering technology, provides an efficient and low-cost innovative path for sludge treatment, and solves the problem of high energy consumption in traditional direct incineration of dewatered sludge. (3) Through the collaborative development of energy-saving treatment devices for high water content sludge and AI prediction models, low-energy clean treatment and intelligent screening of process parameters can be achieved.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an embodiment of the energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to the present invention. Figure 2 This is a schematic diagram of the structure of the energy-saving treatment device for high-moisture sludge based on smoldering and AI prediction according to an embodiment of the present invention; Figure 3 This is a prediction flowchart of an embodiment of the energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction of the present invention; Figure 4 The figure shows a comparison of the mass loss rate results of DTG evolution experiments and XGBoost prediction at 20% and 80% moisture content, where (a) is at 20% moisture content and (b) is at 80% moisture content. Figure 5 Error analysis of DTG evolution experiments and XGBoost predictions at 20% and 80% moisture content, where (a) represents 20% moisture content and (b) represents 80% moisture content; Figure 6 The generalization prediction results of DTG and XGBoost at different heating rates are shown for 20% and 80% moisture content, where (a) is 20% moisture content and (b) is 80% moisture content. Figure Labels 1. Computer; 2. Data acquisition unit; 3. Control unit; 4. Control valve; 5. Gas composition analyzer; 6. Temperature sensor; 7. Heat exchanger tube; 8. Ignition device; 9. Gas flow mixer; 10. Sludge; 11. Ash; 12. Vent; 13. Flow meter; 14. Additional fuel; 15. Gas flow control unit; 16. Air tank; 17. Heating wire. Detailed Implementation

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] Example like Figure 2As shown, the present invention also provides an energy-saving treatment device for high-moisture sludge based on smoldering and AI prediction, including a computer 1, a data acquisition unit 2, a control unit 3, a control valve 4, a gas composition analyzer 5, a temperature sensor 6, a heat exchange tube 7, an igniter 8, an airflow mixer 9, sludge 10, ash 11, a vent 12, a flow meter 13, additional fuel 14, a gas flow control unit 15, an air tank 16, and a heating wire 17. The system includes a data acquisition unit 2 connected to a temperature sensor 6 embedded inside sludge 10 and ash 11; a data acquisition unit 2 connected to a computer 1; a computer 1 connected to a control unit 3; a control unit 3 adjusting a control valve 4 to change the opening of the pipelines for air tank 16 and additional fuel 14; a flow meter 13 installed on the pipelines for air tank 16 and additional fuel 14; both pipelines for air tank 16 and additional fuel 14 connected to an airflow mixer 9; an igniter 8 positioned above sludge 10; an airflow mixer 9 connected to an igniter 8; a vent 12 positioned at the bottom of ash 11, which is located below sludge 10; a heating wire 17 positioned below the vent 12; a gas flow control unit 15 connected to the flow meter 13 to control the supply of additional fuel 14; a heat exchange tube 7 absorbing heat from the high-temperature flue gas at its heat-absorbing end and transferring heat back to the ash at its heat-releasing end; and a gas composition analyzer 5 analyzing the composition of the gas after combustion.

[0023] The workflow is as follows: Air from air tank 16 is metered by flow meter 13 and then enters airflow mixer 9 to provide oxidant for the solid-phase smoldering section and the gas-phase combustion section. Additional fuel 14 is also metered by another flow meter 13 and then enters airflow mixer 9 to assist gas-phase combustion.

[0024] Solid-phase smoldering: Igniter 8 is activated, generating external heat radiation. Heating wire 17 continues to work, transmitting heat through distributed vents 12 at the bottom of ash 11 to the bottom of sludge 10, raising the local temperature of the sludge to the smoldering critical point and triggering the smoldering reaction. In the initial stage, no additional fuel is required; self-sustaining combustion is generated solely by the pyrolysis of air and sludge.

[0025] Temperature sensors 6 are embedded at different depths inside the sludge 10 and ash 11 to collect real-time temperature evolution data of the smoldering front. As the smoldering spreads upward, the ash 11 gradually thickens, and the vent 12 continuously supplies oxygen to maintain the forward advance of the front.

[0026] The sludge 10 undergoes pyrolysis under smoldering conditions, releasing pyrolysis gas containing CH4, CO, and tar vapors. The pyrolysis gas flows upward and enters the airflow mixer 9, where it mixes with air to form a combustible mixture.

[0027] Gas-phase mixing and combustion: In the gas-flow mixer 9, pyrolysis gas, additional fuel 14, and air are fully premixed and ignited by igniter 8 (or rely on high-temperature auto-ignition) before entering the gas-phase combustion section to complete homogeneous oxidation. The supply of additional fuel 14 is automatically adjusted by the gas flow control unit 15 according to the composition of the pyrolysis gas to ensure that the temperature of the gas-phase section is maintained below 1800 K, avoiding thermal NO. x generate.

[0028] The heat exchange tube 7 is placed in the gas phase combustion section to absorb the waste heat of the high-temperature flue gas and transfer the heat back to the bottom of the solid phase smoldering section, near the ash 11. This heat can preheat the unreacted sludge 10, reduce the external ignition energy requirement, achieve two-stage thermal coupling, and improve energy utilization efficiency.

[0029] The treated flue gas enters the gas composition analyzer 5 through the exhaust device, where CO, CO2, and NO are detected online. x The concentration of unburned hydrocarbons was measured to confirm that the toxic components had been completely oxidized and that there were no secondary pollutants. Data acquisition unit 2 collected all signals from temperature sensor 6, gas composition analyzer 5, and gas flow control unit 15, and transmitted them to computer 1 for real-time display, storage, and analysis.

[0030] Computer 1 sends instructions to control unit 3 based on preset programs and real-time feedback data. Control unit 3 adjusts the control valve 4 to change the opening of the pipelines to air tank 16 and additional fuel 14, thereby precisely controlling the flow rates of air and additional fuel. At the same time, gas flow control unit 15 automatically adjusts the supply ratio of additional fuel 14 according to changes in the composition of pyrolysis gas, forming a "monitoring-calculation-adjustment" closed loop to ensure stable, efficient, and low-emission combustion throughout the smoldering and gas-phase combustion processes.

[0031] Verification has shown that smoldering requires only an extremely low heat flux of approximately 25±5 kW / m³. 2 It can be triggered immediately, and the O2 supply is regulated by distributed ventilation (1.5-5 cm / s), enabling smoldering to be self-sustaining without the need for continuous external energy input, with a solid phase burnout rate of more than 95%. The sludge exhibits different thermogravimetric (TG) and differential thermogravimetric (DTG) curves under different moisture contents: the water in the sludge evaporates first at 100℃, and the sludge with high moisture content has a higher DTG evaporation peak. As the temperature rises, the sludge with low moisture content and high moisture content forms a second DTG peak at around 280℃ and 400℃, respectively. This is because the heavy components in it undergo pyrolysis, forming porous coke and pyrolysis gas (CH4, CO, tar vapor); then the coke and O2 undergo heterogeneous oxidation, releasing a large amount of heat (reaction enthalpy of about -28 MJ / kg), and the pyrolysis gas participates in homogeneous oxidation to supplement the heat release (reaction enthalpy of about -2 MJ / kg). After the dual heat release overcomes the heat dissipation and the heat absorption of water evaporation, a self-sustaining cycle of oxidation-driven pyrolysis and pyrolysis supplying fuel is formed.

[0032] Through a two-stage treatment process of solid-phase smoldering and gas-phase combustion, the pyrolysis gas is completely oxidized to CO2 and H2O at a controlled temperature (below 1800 K) in the gas phase section, while inhibiting NO. x The generation and waste heat recycling process in the solid phase section maximizes energy efficiency and significantly saves energy consumption.

[0033] like Figure 1 , Figure 3 As shown, this invention provides an energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction, using smoldering treatment equipment as a carrier, and includes the following steps: S1. Conduct smoldering and heating experiments under different working conditions on sludge samples with different moisture contents to clarify the critical conditions for smoldering self-sustaining under different moisture contents, and obtain thermogravimetric curves and differential thermogravimetric curves in heating experiments under different working conditions.

[0034] First, sludge samples with different moisture contents were prepared and their physicochemical properties were characterized, covering a wide range from low moisture content (20%) to high moisture content (80%). A cone calorimeter (FTT I-Cone Plus) was used as the core experimental setup for smoldering combustion. The cone heater provided uniform radiation intensity to the sample surface, with a radiation intensity range of 25 ± 5 kW / m². 2 The bottom and sides of the sample were insulated with mineral wool, with only the top surface exposed to radiation to ensure one-dimensional heat transfer. The experiment included two modes: continuous radiation and intermittent radiation. Continuous radiation was used to determine the critical ignition heat flux, which is below which smoldering cannot be triggered regardless of heating duration. Intermittent radiation was used to observe whether smoldering was self-sustaining after radiation removal. Based on the experiment, three-zone maps were obtained for smoldering at different moisture contents: no-ignition zone, non-self-sustaining smoldering zone, and self-sustaining smoldering zone.

[0035] Temperature-programmed experiments were conducted using a simultaneous thermal analyzer (NETZSCH STA 409). Thermal gauging (TGA) measurements were performed on the sludge samples at a microscale. Powdered sludge samples were heated in alumina crucibles with micropores. The micropores were used to control the oxygen diffusion rate, simulating the confined oxygen supply microenvironment of a smoldering process, while preventing pressure buildup during pyrolysis. Air velocity was 3 cm / s, heating rates ranged from 5 to 20 K / min in multiple gradients, and the temperature range was from room temperature to above 800 °C. The experiments yielded TG and DTG curves. DTG is defined as the first derivative of TG with respect to time, physically representing the rate of mass loss per unit temperature rise.

[0036] The TG-DTG curve exhibits two characteristic weight loss stages. The first peak (evaporation peak) at approximately 100℃ corresponds to the evaporation of free water and hydrated water, which is a physically endothermic process, and the peak area is proportional to the water content. The second peak (pyrolysis / oxidation peak) at approximately 400℃ corresponds to the pyrolysis and oxidation reactions of organic matter (taking 80% water content as an example), which is a coupled process of chemical decomposition and partial exothermic reaction.

[0037] The corresponding physicochemical equation is: (Water evaporates, absorbing heat, approximately 100℃); (Pyrolysis of organic matter, endothermic, triggered at approximately 250°C); (Coke oxidation, exothermic, triggered at ≥350℃); (Pyrolysis gas oxidation, exothermic); The two peaks wax and wane depending on the moisture content. At low moisture content, the evaporation peak is minimal, and heat is rapidly used to heat the organic matter, resulting in a high and narrow pyrolysis / oxidation peak that is easily self-sustaining. At high moisture content, the evaporation peak increases significantly, as a large amount of water evaporates, consuming heat and diluting oxygen by occupying pores with water vapor. This creates a dual inhibition of endothermic competition and mass transfer obstruction, causing the pyrolysis / oxidation peak to weaken significantly and become flat. By increasing the distributed ventilation rate to force the removal of water vapor and replenish oxygen, the inhibition of smoldering by high moisture content can be effectively compensated.

[0038] S2. Measure the changes in flue gas composition, temperature field, and mass during the multi-condition smoldering test to obtain data on flue gas composition, temperature field, and mass changes. Based on experiments using a cone calorimeter, real-time online measurements of flue gas composition, temperature field, and mass changes were performed. Flue gas composition was measured using a non-dispersive infrared absorption gas analyzer. CO2 exhibits a strong infrared absorption band at 4.26 μm, and CO shows characteristic absorption at 4.67 μm. When infrared light passes through the gas being measured, quantification was performed according to Beer-Lambert's law. ; In the formula, A Absorbance; ε The molar absorptivity; b Optical path length; c Gas concentration; I 0 and I The incident and transmitted light intensities are represented by the absorbance. Absorbance is directly proportional to concentration, thus enabling real-time quantitative measurement of CO and CO2.

[0039] After being collected by a fume hood, the flue gas is extracted by a sampling pump at a flow rate of approximately 1.5 L / min and sent to an analyzer to obtain the CO and CO2 emission fluxes and the CO / CO2 ratio. Smoldering is incomplete combustion, characterized by a CO / CO2 ratio of approximately 0.1 to 0.2, significantly higher than flaming combustion. The formation pathway of toxic combustible components in the flue gas is as follows: (Pyrolysis gaseous products); (Incomplete oxidation of carbon); (Main channel for CO oxidation); CO is an incomplete oxidation product of carbon, while CH4, tar vapor, and polycyclic aromatic hydrocarbons are pyrolysis products of organic matter. The main oxidation pathway for CO depends on OH radicals. All of these components must be completely oxidized to CO2 and H2O in the gas-phase combustion stage.

[0040] Temperature measurement employs a K-type thermocouple array, based on the Seebeck effect—the thermoelectric potential generated when there is a temperature difference between the two ends of two metallic conductors. ; Nickel-chromium / nickel-silicon K-type thermocouple Approximately 41 μV / ℃, with a temperature range of -200 to 1250℃, covering the entire smoldering temperature range. Thermocouples are arranged at three depths below the sample surface along the central axis: 0 mm, 10 mm, and 20 mm. Surface thermocouples are covered with a thin layer of sludge to ensure measurement of the solid phase temperature, while deep thermocouples monitor frontal advance. Signals are acquired at 5-second intervals using a Midi Logger GL840. The frontal advance rate is determined by the displacement of the characteristic temperature isotherm. ; In the formula, The forward propulsion rate; This represents the displacement of the front between the two thermocouples; This corresponds to the time interval.

[0041] Mass changes were recorded in real time using a precision balance (±0.1 mg). The mass loss rate and heat release parameters were defined as follows: ; ; ; In the formula, For mass loss rate; The rate of heat release; For the first The enthalpy of the first reaction is negative when it is exothermic. The initial mass; This is the cumulative heat release. PMLR Take | MLR The maximum value corresponds to the peak temperature range of the DTG. For the first The reaction rate of a step reaction.

[0042] S3. Based on the three-step principle of smoldering reaction, and using the obtained thermogravimetric curve and differential thermogravimetric curve, calculate the kinetic parameters of smoldering reaction, and optimize the kinetic parameters using the least squares objective function.

[0043] During smoldering, the high-moisture sludge undergoes three characteristic stages. The first stage is water evaporation: free water and hydrated water in the sludge vaporize endothermally at approximately 100°C. The weight loss in this stage depends on the initial moisture content. The heat consumed in water evaporation delays the triggering of subsequent reactions and reduces overall thermal efficiency. The second stage is organic matter pyrolysis: the remaining high-moisture solid phase is heated to approximately 400°C under anaerobic conditions, causing organic macromolecules to break down into porous coke and gaseous volatiles. The volatiles mainly consist of methane, carbon monoxide, and tar vapor. This stage is endothermic and kinetically described by Arrhenius law. The third stage is oxidation: the coke undergoes heterogeneous oxidation with oxygen diffused into the sludge pores, producing ash, carbon dioxide, and a small amount of carbon monoxide. The combustible components in the pyrolysis gas are homogeneously oxidized into carbon dioxide and water in the gas phase. The combined heat release from the two oxidation pathways constitutes the total heat production of the system. When the heat release exceeds the sum of environmental heat dissipation and the heat endothermic from water evaporation, smoldering achieves self-sustaining propagation without the need for continuous external energy input.

[0044] Based on the macroscopic experiments in S1 and S2, and using the two core curves (TG and DTG) obtained in step S1, the conversion rate α is defined based on the TG data to normalize the reaction process. ; In the formula, For the initial mass, For instantaneous sample mass, The final sample mass; ∈[0,1], =0 indicates that the reaction has not started. =1 indicates that the reaction is complete.

[0045] The reaction rate in each of the three steps follows Arrhenius's law, meaning the reaction rate constant is exponentially related to temperature. The reaction rate equations for pyrolysis, sludge oxidation, and carbon oxidation are as follows: ; ; ; In the formula, r 1. r 2. r 3 represents the reaction rates of pyrolysis, sludge oxidation, and carbon oxidation, respectively; A 1. A 2. A 3 represents the pre-exponential factor for pyrolysis, sludge oxidation, and carbon oxidation reactions; E 1. E 2. E 3 represents the activation energy for pyrolysis, sludge oxidation, and carbon oxidation reactions; R This is the universal gas constant;T The reaction temperature; , , These represent the mass fractions of sludge, coke, and oxygen, respectively, dimensionless. denoted as the reaction order.

[0046] The activation energy was estimated without a model using the Kissinger method, based on the shift in the peak DTG temperature at different heating rates. The basic equation of the Kissinger method is: ; In the formula, For the heating rate, This represents the peak temperature of the DTG curve. Represents the square of the peak temperature, in right Plot the graph, obtain the activation energy E from the slope of the straight line, and obtain the pre-exponential factor from the intercept. A .

[0047] Based on the above reaction rates, the evolution of the mass fractions of the three solid components, namely sludge, coke, and ash, over time is described by a set of mass conservation equations: ; ; ; In the formula, , , These are the stoichiometric coefficients of the solid-phase products in the pyrolysis, sludge oxidation, and carbon oxidation reactions. Sludge is consumed in both pyrolysis and oxidation steps; coke has two sources of formation, one from the pyrolysis step. and the products generated in the sludge oxidation process And a consumption pathway, namely the carbon oxidation step. Ash is generated solely from the carbon oxidation process.

[0048] The kinetic parameters of the pyrolysis, sludge oxidation, and carbon oxidation reactions were optimized using a least-squares objective function to obtain the kinetic parameters of each reaction; the least-squares objective function is as follows: ; In the formula, The parameter vector to be optimized includes the pre-exponential factors, activation energies, and stoichiometric coefficients for pyrolysis, sludge oxidation, and carbon oxidation reactions. N For the number of experimental data points, These are experimental values. For the corresponding simulated values, This is the error normalization scaling factor.

[0049] S4. Organize and preprocess all experimental data obtained in steps S1 to S3, construct a multidimensional database, and divide the database into training, validation, and test sets in a 7:2:1 ratio. The input feature vector includes TG curve data and DTG curve data under different moisture contents, as well as operating parameters such as heating rate and radiation intensity. The TG curve data includes mass loss, and the DTG curve data includes mass loss reaction rate. The output target vector Y includes mass loss and mass loss reaction rate.

[0050] The data processing includes organizing the TG / DTG curve data of sludge with different moisture contents and their correspondence with operating parameters (moisture content, heating rate, etc.), and discretizing them into structured samples according to temperature points.

[0051] Preprocessing includes data standardization and feature correlation calculation. Data standardization uses Z-score normalization: ; In the formula, The characteristic mean, The standard deviation is used to ensure that all input features fall within the same order of magnitude.

[0052] The training database can be represented as a feature matrix: ; In the formula, N For the sample size, D For feature dimensions.

[0053] Feature correlation is calculated using the Pearson correlation coefficient matrix: ; In the formula, Representation of features With features The Pearson correlation coefficient between them, with a value range of [-1, 1]. Representation of features standard deviation Representation of features The standard deviation is used to assess the linear correlation between features and to eliminate redundant features.

[0054] To compensate for the insufficient coverage of experimental conditions, batch calculations were performed for conditions with different moisture contents. The simulated TG / DTG curves and kinetic parameters were added to the database as expanded training samples. All data were randomly divided into training, validation, and test sets according to a set ratio. The training set was used for model parameter learning, the validation set was used for hyperparameter tuning, and the test set was used for final generalization ability evaluation.

[0055] S5. Construct a prediction model. Train the prediction model using the training set, optimize by minimizing the least squares objective function, control prediction bias, and fine-tune the hyperparameters of the trained prediction model using the validation set to obtain a high-generalization prediction model.

[0056] XGBoost is an ensemble learning algorithm based on gradient boosting decision trees. Its core idea is to gradually add CART regression trees as weak learners, train a new tree along the negative gradient direction of the previous round residual in each iteration, and then sum the prediction results of all trees in a weighted manner to obtain the final output.

[0057] The target loss function of the prediction model consists of two parts: a training loss term and a regularization term. The loss function is as follows: ; The regularization terms are as follows: ; In the formula, For the first The prediction loss for each sample is calculated, and the mean squared error is used for the regression task. These are the model's predicted values. This represents the actual experimental value. K The total number of trees, For the k-th regression tree, T The number of leaf nodes in the tree. w The leaf weight vector, The penalty coefficient for the number of leaves. for L 2. Regularization coefficient, both are used together to suppress overfitting and enhance the model's generalization ability.

[0058] During training, the prediction model employs a second-order Taylor expansion to approximate the loss function to accelerate convergence and incorporates column sampling and shrinkage mechanisms. Model hyperparameters include the learning rate. Maximum tree depth Minimum leaf weight , L 2 Regularization coefficient The optimization selection is performed through grid search and cross-validation on the validation set, with the least squares objective function. M Minimization is the optimization direction, aiming to make the overall deviation between the fitted curve and the experimental data less than 5% across the entire temperature range.

[0059] S6. The generalization ability of the high generalization prediction model is evaluated using a test set. Based on the evaluation, the prediction model is used to predict different combinations of working conditions for sludge with high water content. Based on the prediction results, a smoldering self-sustaining criticality prediction spectrum is established.

[0060] The generalization ability of the trained prediction model is evaluated using a test set. Evaluation metrics include the coefficient of determination R0.2 The root mean square error (RMSE) and the point-by-point deviation between the TG / DTG experimental curve and the fitted curve of the prediction model are considered. The coefficient of determination is defined as: ; The root mean square error is: ; In the formula, The coefficient of determination measures the goodness of fit; the closer to 1, the better. Let be the experimental true value of the i-th sample; This is the model prediction value for the i-th sample; This is the mean of all true values. RMSE is the root mean square error, which measures the average deviation between the predicted and true values. The total number of samples.

[0061] By comparing the prediction results of different working conditions in the test set with the experimental data, we can verify whether the model has reliable generalization prediction ability for working conditions that were not trained.

[0062] After verifying the generalization ability of the prediction model, the prediction model was used to perform batch inference predictions on a large number of working condition combinations for which no experiments were conducted, and a smoldering self-sustaining critical prediction spectrum of radiation intensity-ignition time-moisture content was established.

[0063] S7. Based on the established smoldering self-sustaining criticality prediction map, the prediction model is verified and optimized. Based on the optimized prediction model, the global operating space of high water content sludge is predicted to obtain the optimal operating condition map.

[0064] A novel operating condition—one not included in the training, validation, and test sets (i.e., a moisture content not found in the database)—was selected for supplementary validation experiments. The TG / DTG experimental curves under this new condition were compared point-by-point with the fitted curves of the prediction model, and the average deviation across the entire temperature range was calculated. The prediction deviation was defined as: ; In the formula, ε Prediction bias is expressed as a percentage. C To verify the total number of experimental data points, For the first The model prediction value for each data point. This is a supplementary experimental measurement for the j-th data point.

[0065] The model prediction results also provide confidence intervals. In the formula The standard deviation of the model's predictions on the test set. z is the confidence coefficient.

[0066] If the average deviation between the prediction and the experiment exceeds the preset accuracy threshold, the experimental data for that working condition is added to the training database, the prediction model hyperparameters are iteratively optimized, and the prediction model is retrained until the prediction deviation of the prediction model for the new working condition meets the accuracy requirements, at which point the iteration stops.

[0067] After the model accuracy meets the requirements, the converged prediction model is used to scan and predict the global operating space of high-moisture-content sludge, generating an optimal operating condition map covering different moisture contents and radiation intensities. This map directly provides the optimal radiation intensity for sludge with different moisture contents, as well as the corresponding predicted burnout rate and reaction time, outputting a sludge smoldering treatment process parameter table. This parameter table can directly guide the setting of operating parameters for the solid-phase smoldering section and the gas-phase combustion section in the smoldering treatment equipment, providing AI decision-making basis for large-scale sludge treatment practical engineering applications.

[0068] The XGBoost prediction model of this invention achieves high-precision prediction under low moisture content conditions. For example... Figure 4 As shown in (a), under the condition of 20% low moisture content, the experimental DTG curve exhibits a single sharp pyrolysis / smoldering peak, located at approximately 300-350℃, with a peak value close to -10% / min. In the high-temperature region (>600℃), the curve flattens out and approaches zero. The XGBoost predicted value fitting curve highly overlaps with the experimental curve, especially showing excellent agreement in the peak range, with only slight differences at local minor fluctuations. Figure 5 (a) Error analysis shows that most of the scattered points are closely distributed near the ideal diagonal and all fall within the ±20% error band. The points are denser near the origin, while the scattered points in the peak area (actual value approximately -8 to -10% / min) are slightly dispersed but still remain within the ±20% band. This indicates that XGBoost has small prediction errors and high accuracy under low moisture content. The mechanism is as follows: under low moisture content conditions, the water content in solid waste is limited, the heat absorption of evaporation is minimal, and the TG-DTG curve is dominated by only a single organic pyrolysis / oxidation reaction. The reaction path is relatively simple, and the DTG characteristic shows a concentrated and sharp single-peak structure. XGBoost captures the temperature-weight loss rate mapping relationship of this single reaction path through the decision tree splitting mechanism, and the residual iteration converges sufficiently, thus resulting in extremely high prediction accuracy. Taking the 20% low moisture content condition as a unified reference benchmark, the model's prediction performance under this condition can be used as a comparative baseline for the generalization ability under subsequent complex high moisture content conditions.

[0069] The XGBoost prediction model of this invention can accurately capture the bimodal competition characteristics of DTG under high moisture content conditions. For example... Figure 4As shown in (b), under the condition of 80% high moisture content, the experimental DTG curve exhibits a significant bimodal structure: the first peak is located at approximately 100-150℃, with an amplitude of approximately -3 to -4% / min, corresponding to the endothermic evaporation of free water and hydrated water; the second peak is located at approximately 300-350℃, with an amplitude of approximately -5 to -6% / min, corresponding to the pyrolysis and smoldering reaction of organic matter. Compared to the low moisture content condition, the peak value of the second peak decreases from approximately -10% / min to approximately -5 to -6% / min, a decrease of approximately 40%-50%, fully highlighting the inhibitory effect of high moisture content on the smoldering reaction. The XGBoost predicted value fitting curve can accurately identify the positions of the two characteristic peaks and generally follows the experimental curve trend, but there are certain deviations in the peak shape of the first peak and the amplitude of the second peak; the predicted value is slightly higher than the experimental value or the peak shape is slightly wider. Figure 5 (b) Error analysis shows that the scatter plot distribution is relatively more dispersed. In the regions with true values ​​of -1 to -3% / min (corresponding to the evaporation peak) and -4 to -6% / min (corresponding to the smoldering peak), some points slightly exceed the ±20% lower or upper boundary, but most scatter plots remain within the ±20% band, and the overall prediction error is acceptable. The mechanism is that under high moisture content conditions, the two reaction pathways of water evaporation and organic matter pyrolysis / oxidation are coupled and compete with each other. That is, water evaporation consumes a lot of heat, delaying the triggering of subsequent reactions. At the same time, the water vapor generated occupies the pores of solid waste and dilutes the oxygen concentration, resulting in a dual inhibition of endothermic competition and mass transfer hindrance. XGBoost needs to learn the temperature-weight loss rate mapping and their interaction relationship of the two reaction pathways in the bimodal feature space at the same time. The feature dimension increases and the data distribution becomes more complex. The fitting accuracy decreases slightly in some areas due to the cumulative error of the second-order Taylor expansion, but the overall physical rationality of the model is still reliable.

[0070] The XGBoost prediction model of this invention demonstrates a physical generalization ability under multiple heating rates. For example... Figure 6 As shown, at different heating rates (5℃ / min, 10℃ / min, 15℃ / min), XGBoost accurately reproduced the thermal hysteresis law that the higher the heating rate, the larger the absolute value of the DTG peak and the shift towards the high-temperature region. Under low moisture content conditions ( Figure 6 (a) The predicted peak value at 5℃ / min is approximately -6 to -7% / min, at 10℃ / min it is approximately -10% / min, and at 15℃ / min it is approximately -12 to -13% / min. The predicted curve at 10℃ / min almost coincides with the actual curve, verifying the model's accuracy within the training conditions. Under high moisture content conditions ( Figure 6(b) The changing trends of the two characteristic peaks with the heating rate were also accurately captured—the amplitudes of both peaks decreased at 5℃ / min, and increased at 15℃ / min. The moisture evaporation peak lagged slightly, while the smoldering peak shifted significantly to the right and its amplitude increased. The mechanism is that the change in heating rate essentially alters the rate of heat accumulation and the reaction kinetic response within the solid waste—the higher the heating rate, the greater the heat input per unit time, the faster the reaction rate, and the larger the DTG peak value. At the same time, the thermal hysteresis effect causes the characteristic temperature to shift towards the high-temperature region. XGBoost can accurately reproduce the above physical laws at different heating rates, and maintains prediction accuracy even at heating rates not used in training. This indicates that the model automatically learns the physical essence of the smoldering reaction through the decision tree splitting mechanism, namely the Arrhenius kinetic constraint relationship between temperature, heating rate, and reaction rate, rather than an overfit to a specific distribution of the training data. Therefore, it demonstrates reliable generalization prediction ability when extrapolating in both the moisture content and heating rate dimensions.

[0071] Therefore, this invention adopts the above-mentioned energy-saving treatment method and device for high-moisture sludge based on smoldering and AI prediction. Using smoldering treatment equipment as a carrier, it directly feeds high-moisture sludge without pre-drying. The system can be distributed and deployed in small and medium-sized workstations, achieving low-energy clean treatment through waste heat recycling. At the same time, the XGBoost model trained with limited TGA experimental data has strong generalization ability and can predict TG / DTG curves and treatment effect indicators for any moisture content. The prediction deviation is controlled within 5%, breaking through the efficiency bottleneck of traditional methods that rely on condition-by-condition experiments. This generalization model provides universal AI decision support for rapid optimization and intelligent control of process parameters in future multi-condition scenarios.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An energy-saving treatment method for high-moisture sludge based on smoldering combustion and AI prediction, characterized in that, Includes the following steps: S1. Conduct smoldering and heating experiments under different working conditions on sludge samples with different moisture contents, clarify the critical conditions for smoldering self-sustaining under different moisture contents, and obtain thermogravimetric curves and differential thermogravimetric curves in heating experiments under different working conditions. S2. Measure the changes in flue gas composition, temperature field, and mass during the smoldering experiment to obtain data on flue gas composition, temperature field, and mass changes; S3. Based on the three-step smoldering reaction principle, and using the obtained thermogravimetric curves and differential thermogravimetric curves, calculate the kinetic parameters of the smoldering reaction, and optimize the kinetic parameters using the least squares objective function. S4. Organize and preprocess the experimental data obtained from steps S1 to S3, construct a multidimensional database, and divide the multidimensional database into a training set, a validation set, and a test set. S5. Construct a prediction model. Train the prediction model using a training set, with the optimization direction being minimizing the least squares objective function to control prediction bias. Use a validation set to fine-tune the hyperparameters of the trained prediction model to obtain a high-generalization prediction model. The prediction model is constructed using the XGBoost extreme gradient boosting algorithm. The input feature vector includes thermogravimetric curve data, differential thermogravimetric curve data, heating rate, and radiation intensity under different moisture content conditions. The thermogravimetric curve data includes mass loss, and the differential thermogravimetric curve data includes mass loss reaction rate. The output target vector includes mass loss and mass loss reaction rate. The loss function consists of a loss term and a regularization term. S6. The generalization ability of the high generalization prediction model is evaluated using a test set. Based on the evaluated prediction model, predictions are made for different combinations of working conditions for sludge with high water content. A smoldering self-sustaining critical prediction spectrum is established based on the prediction results. The smoldering self-sustaining critical prediction spectrum is established based on radiation intensity, ignition time and water content. S7. Based on the established smoldering self-sustaining criticality prediction map, the prediction model is validated and optimized. The optimized prediction model is then used to predict the global operating space of high-moisture-content sludge, yielding the optimal operating condition map, specifically including: A new operating condition not included in either the training or test set was selected for supplementary verification experiments; The thermogravimetric curves and differential thermogravimetric curves under the new operating conditions are compared point by point with the fitted curves of the predicted values ​​from the prediction model. The average deviation over the entire temperature range is calculated, and the deviation is defined as follows: ； In the formula, ε Prediction bias is expressed as a percentage. C To verify the total number of experimental data points, For the first The model prediction value for each data point. This is a supplementary experimental measurement for the j-th data point; If the average deviation between the prediction and the experiment exceeds the preset accuracy threshold, the experimental data for that working condition is added to the training database, the prediction model hyperparameters are iteratively optimized, and the prediction model is retrained until the prediction deviation of the prediction model for the new working condition meets the accuracy requirements, at which point the iteration stops.

2. The energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to claim 1, characterized in that: Operating conditions include radiation intensity and heating rate.

3. The energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to claim 2, characterized in that, Step S3 specifically includes: S31. Define the conversion rate based on thermogravimetric curve data, normalize the reaction process, and determine the reaction state; S32. Based on the three-step smoldering reaction principle, estimate the kinetic parameters of pyrolysis, sludge oxidation and carbon oxidation reactions based on Arrhenius law and the peak value of differential thermogravimetric curves. The three-step smoldering reaction includes water evaporation, pyrolysis and oxidation, and oxidation includes sludge oxidation and carbon oxidation. S33. The kinetic parameters of the pyrolysis, sludge oxidation and carbon oxidation reactions are optimized and solved using the least squares objective function to obtain the kinetic parameters of each reaction.

4. The energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to claim 3, characterized in that, The kinetic parameters include the pre-exponential factor and activation energy.

5. The energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to claim 3, characterized in that: The least squares objective function is: ； In the formula, The parameter vector to be optimized includes the pre-exponential factors, activation energies, and stoichiometric coefficients for pyrolysis, sludge oxidation, and carbon oxidation reactions. N For the number of experimental data points, These are experimental values. For the corresponding simulated values, This is the error normalization scaling factor.

6. The energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to claim 1, characterized in that: Evaluation metrics include the coefficient of determination, root mean square error, and point-by-point deviation between the thermogravimetric and differential thermogravimetric experimental curves and the fitted curves of the prediction model.

7. An energy-saving treatment device for high-moisture sludge based on smoldering and AI prediction, comprising the energy-saving treatment method for high-moisture sludge based on smoldering and AI prediction according to any one of claims 1-6, characterized in that, The system includes a computer, data acquisition unit, control unit, control valve, gas composition analyzer, temperature sensor, heat exchange tube, igniter, airflow mixer, vent, flow meter, additional fuel, gas flow control unit, air tank, and heating wire. The data acquisition unit is connected to the temperature sensor, which is embedded inside the sludge and ash. The data acquisition unit is connected to the computer, which in turn is connected to the control unit. The control unit adjusts the opening of the air tank and additional fuel pipelines by regulating the control valve. The flow meter is installed on the air tank and additional fuel pipelines, both of which are connected to the airflow mixer. The igniter is located above the sludge, and the airflow mixer is connected to the igniter. The vent is located at the bottom of the ash, which is located below the sludge. The heating wire is located below the vent. The gas flow control unit is connected to the flow meter and controls the supply of additional fuel. The heat exchange tube absorbs the waste heat from the high-temperature flue gas at its heat-absorbing end and transfers heat back to the ash at its heat-releasing end. The gas composition analyzer analyzes the composition of the gases after combustion.

Images