A machine learning-based sludge-derived synthetic fuel formulation optimization method

Abstract

The application discloses a sludge-derived fuel formula optimization method based on machine learning, which comprises the following steps: collecting municipal sludge, industrial sludge and auxiliary biomass fuel basic physicochemical property parameters, obtaining different proportion combustion characteristics and pollutant release parameters based on TG-FTIR and TG-MS, and supplementing investigation data to form a sludge-derived fuel data set; taking the low calorific value of the formed fuel, the comprehensive combustion index, the combustion stability index, the release amounts of NO x , SO2 and HCl as target indexes, establishing an optimal machine learning prediction model, then forming a graphical user interface (GUI) for sludge-derived fuel combustion characteristic prediction, and realizing rapid operation and display; further combining a multi-objective optimization algorithm, obtaining a recommended formula interval of the sludge-derived fuel meeting the calorific value, combustion stability and pollutant emission constraints; and finally, based on the optimized formula, carrying out parameter optimization of the forming process, and providing basic guidance for industrialized production of sludge forming fuel preparation.

Description

Technical Field

[0001] This invention relates to the field of solid waste resource utilization and clean fuel preparation technology, and in particular to a method for optimizing sludge-derived fuel formulations based on machine learning. Background Technology

[0002] Sludge incineration has attracted much attention due to its high volume reduction and harmless treatment effects. However, municipal sludge has a low calorific value due to its high water content and high ash content. Blending it with coal-fired power plants has a significant impact on the load.

[0003] Therefore, by drying, blending, molding, and baking municipal sludge with other high-calorific-value fuels (coal, biomass, organic solid waste, etc.) to produce granular or cylindrical sludge-derived fuels, the energy density of sludge-based fuels can be improved and stable combustion can be achieved.

[0004] However, the composition of sludge raw materials is complex and fluctuates significantly. Different batches of sludge have large differences in moisture content, ash content, volatile matter, fixed carbon, calorific value, nitrogen, sulfur and chlorine content and heavy metal content, which makes the design of sludge-derived fuel formulations quite difficult.

[0005] Existing methods for preparing sludge-derived fuels mostly rely on manual experience and a small number of experiments to determine the formula ratio. Typically, sludge, biomass, coal powder, sulfur fixatives, and binders are mixed and molded in a fixed ratio.

[0006] Chinese patent CN110835568B discloses a method for preparing sludge-derived fuel for hazardous waste disposal, specifically as follows:

[0007] (a) Receiving of sludge-type hazardous waste, industrial solid waste, auxiliary fuel, conditioning powder, desulfurization and deacidification agent powder and binder powder;

[0008] Sludge-type hazardous waste and industrial solid waste are transported to the treatment plant by transport vehicles; non-powder sludge-type hazardous waste is weighed on a weighbridge and then unloaded into the sludge temporary storage warehouse of the treatment plant, and non-powder industrial solid waste is weighed on a weighbridge and then unloaded into the industrial solid waste temporary storage warehouse of the treatment plant; the remaining powdery sludge powder and industrial solid waste fine powder are weighed on a weighbridge and then directly unloaded into the sludge powder silo and industrial solid waste powder silo of the treatment plant.

[0009] The purchased auxiliary fuel is transported to the disposal plant by transport vehicles and unloaded into the auxiliary fuel warehouse of the disposal plant after being weighed by a weighbridge. The auxiliary fuel is coal, straw or a mixture of both. Coal is unloaded into the coal warehouse and straw is unloaded into the straw warehouse.

[0010] The purchased regulator powder, desulfurization and deacidification agent powder, and binder powder are transported to the treatment plant by transport vehicles, and after being weighed by a weighbridge, they are unloaded into the regulator powder silo, desulfurization and deacidification agent powder silo, and binder powder silo of the treatment plant, respectively.

[0011] (ii) Pretreatment;

[0012] (1) Drying;

[0013] The sludge in the sludge temporary storage warehouse is transported to the sludge dryer by transfer and conveying machinery to dry the wet sludge into sludge powder with a moisture content of 15-40%; industrial solid waste is dried together with wet sludge, or a separate industrial solid waste dryer is set up for drying and dewatering.

[0014] (2) Broken;

[0015] The auxiliary fuel in the auxiliary fuel depot is transported to the auxiliary fuel crusher via a transfer conveyor to crush the auxiliary fuel into fine powder less than 2mm. When coal is used as the auxiliary fuel, a coal crusher is used; when straw is used as the auxiliary fuel, a straw crusher is used.

[0016] The industrial solid waste in the temporary storage warehouse is transported to the industrial solid waste crusher by transfer and conveying machinery, and the industrial solid waste is crushed into fine industrial solid waste powder smaller than 2mm.

[0017] (3) Temporarily store the ingredients;

[0018] The dried sludge powder is transported to the sludge powder silo, the crushed auxiliary fuel powder is transported to the auxiliary fuel powder silo, and the crushed industrial solid waste powder is transported to the industrial solid waste powder silo.

[0019] (III) Compatibility;

[0020] SDF (Self-Detergent-Based Fuel) feedstock consists of main and auxiliary materials. Sludge powder is the main material, while auxiliary materials include fine auxiliary fuel powder, fine industrial solid waste powder, conditioner powder, desulfurization and deacidification agent powder, binder powder, and conditioning water. The blending process is as follows: The sludge powder and fine industrial solid waste powder are analyzed by the treatment plant's laboratory to determine their specific components and calorific value. Then, based on the respective quantities to be treated, the proportions of sludge powder, fine industrial solid waste powder, and fine auxiliary fuel powder in the first blend are calculated. Based on the specific gravity of the pollutants in the first blend, the corresponding consumption of desulfurization and deacidification agent powder is calculated, and the relative amounts are then calculated. The required amount of regulator powder and the amount of binder added to achieve the required strength for subsequent molding are calculated to obtain the proportions of main and auxiliary materials in the secondary formulation. Based on the moisture content of the secondary formulation, the amount of conditioning water is calculated to complete the proportions of main and auxiliary materials in the tertiary formulation. The scientific validity of the tertiary formulation is verified, and the main and auxiliary materials are fine-tuned to obtain the accurate production formula. Specifically, verifying the scientific validity of the tertiary formulation involves verifying the strength, calorific value, Ca / S ratio, ash melting point, and slag basicity of the SDF product.

[0021] Metering devices are installed below the sludge powder silo, auxiliary fuel powder silo, industrial solid waste powder silo, regulator powder silo, desulfurization and deacidification agent powder silo, and binder powder silo. According to the production formula, the mixed materials in the main material silo and various auxiliary material silos are measured separately and then transported to the buffer silo by transfer conveyor machinery.

[0022] (iv) Mixing;

[0023] The material in the buffer silo is transported to the mixer by a transfer conveyor to be fully mixed and stirred evenly; at the same time, metered conditioning water is sprayed into the mixer to adjust and control the moisture content; after the SDF molding material is mixed and conditioned, it is transported to the transition silo by a transfer conveyor; the moisture content of the SDF molding material after being fully mixed and conditioned is controlled at 15-20%;

[0024] (V) Shaping;

[0025] The SDF molding material in the transition chamber is transported to the molding machine by a transfer conveyor, where it is squeezed and mixed to become molded SDF wet material. The freshly molded SDF wet material is screened by a vibrating screen to remove SDF fragments. The SDF fragments are returned to the mixer by a transfer conveyor to participate in molding again. The molded SDF wet material is rolled into the transfer conveyor through the vibrating screen outlet and sent to the drying step by the transfer conveyor.

[0026] (vi) Drying;

[0027] The molded SDF wet material is packed in ton bags or directly unloaded into the transfer site and reused after natural drying; or the molded SDF wet material is directly transported to the dryer for forced thermal drying via transfer conveyor machinery; the moisture content of the dried SDF product is less than 10%, and it meets the three compatibility scientific verification indicators of SDF products.

[0028] The conditioning water is derived from process wastewater, reclaimed water, softened water, and tap water; one or more conditioning water tanks are provided, the bottom outlet of the conditioning water tank is connected to the conditioning water pump through a pipeline, the outlet of the conditioning water pump is connected to the conditioning water metering device through a pipeline, and the outlet of the conditioning water metering device is connected to the mixer through a pipeline; the conditioning water tank holds various types of conditioning water or binder solutions.

[0029] The scientific validity of the three compatibility checks were verified, specifically by checking the strength, calorific value, Ca / S ratio, ash fusion point, and slag basicity of the SDF product. The strength of the SDF product was verified to be greater than the preset minimum value. The calorific value of the SDF product was verified based on the furnace type characteristics of the high-temperature solid waste heat treatment equipment, requiring a calorific value greater than 1800 kcal / kg when a gasification melting furnace is selected. The Ca / S ratio of the SDF product was verified to be Ca / S = 1–3. The ash fusion point of the SDF product was verified based on the furnace type characteristics of the high-temperature solid waste heat treatment equipment, requiring an ash fusion point less than 1400℃ when a gasification melting furnace is selected. The slag basicity of the SDF product was verified to be total basicity R = 0.3–1.5.

[0030] Chinese patent CN113526815B discloses a method for preparing sludge-derived clean fuel, specifically including the following steps: S1, pretreating sludge-containing wastewater to remove odor and heavy metal ions, obtaining wastewater A; S2, adding a wastewater modifier to wastewater A obtained after step S1, mixing evenly, and aging for 1-2.5 hours to obtain wastewater B; wherein the wastewater modifier comprises component A and component B, and the mass ratio of wastewater A, component A, and component B is 100:0.5:0.25; wherein component A is composed of the following raw materials in parts by weight: hypochlorite 5-20 parts, dichromate 8-20 parts, chromate 8-20 parts, copper nitrate 20-35 parts, calcium nitrate 20-38 parts, sodium nitrate 25-41 parts, and component B is composed of the following raw materials in parts by weight: Bacillus subtilis 3-6 parts, lactobacillus 1-4 parts, Aspergillus niger 1.5- 5 portions; S3. Introduce ozone micro-nano bubbles into wastewater B obtained after treatment in step S2. Aerate the wastewater B by introducing ozone micro-nano bubbles to ensure full contact between ozone and wastewater B. Simultaneously, treat wastewater B with ozone micro-nano bubbles using ultrasound. The strong oxidizing properties of ozone micro-nano bubbles destroy extracellular polymers and cell walls in the wastewater. After treatment with ozone micro-nano bubbles and ultrasound, separate the oil phase contained in wastewater B to obtain wastewater C; S4. Treat wastewater C obtained after treatment in step S3 using high-voltage electronic pulse technology, and then mechanically dewater it until the water content is below 30% to obtain sludge; S5. Add auxiliary fuel to the sludge obtained after treatment in step S4. After thorough mixing, allow the mixture to dry naturally until the water content is below 20%. Then, press the mixture into shape to obtain sludge-derived clean fuel; wherein, the mass ratio of sludge obtained after treatment in step S4 to the auxiliary fuel is 25-80:10-45.

[0031] It is evident that both methods use sludge as the main material, and then add auxiliary materials according to experimental experience to formulate, mix, mold, and dry the sludge-based briquettes to produce sludge-based briquettes.

[0032] However, this type of method has the following shortcomings:

[0033] 1) The formulation has poor adaptability to fluctuations in sludge quality. When the sludge moisture content, ash content or calorific value changes, the strength, combustion performance and pollutant release of the briquetted fuel are prone to fluctuations.

[0034] 2) Formulation development relies on a large number of repeated experiments, which is time-consuming and costly, making it difficult to quickly obtain the optimal formulation that meets multiple constraints.

[0035] 3) Existing methods usually only consider calorific value or molding strength, and rarely consider multiple objectives such as combustion characteristics and stability, and pollutant emissions at the same time;

[0036] 4) The lack of a feedback correction mechanism based on historical experimental and operational data makes it difficult to achieve continuous optimization of the formulation and engineering scale-up.

[0037] In recent years, machine learning methods have shown great potential in modeling complex nonlinear systems, optimizing material formulations, co-processing solid waste, and predicting combustion performance. Introducing machine learning into the design of sludge-derived briquettes can establish a mapping relationship between raw material characteristics, formulation ratios, molding process parameters, and fuel performance. This enables rapid prediction of the calorific value, strength, combustion characteristics, pollutant release, and cost of the briquettes, and allows for the acquisition of optimal formulations that meet engineering constraints through multi-objective optimization.

[0038] However, current machine learning-based formulation optimization methods for sludge-derived briquette fuels are still imperfect, and there is a lack of a systematic predictive optimization method that addresses sludge quality fluctuations, additive fluctuations, briquette fuel product performance, and engineering application constraints.

[0039] Therefore, in order to solve the problems of existing sludge-derived fuel formulations relying on experience, poor adaptability to fluctuations in sludge quality and biomass additives, insufficient multi-objective synergistic optimization, and long experimental development cycles, it is an urgent technical problem to be solved by those skilled in the art to provide a machine learning-based sludge-derived fuel formulation optimization method to improve the accuracy, stability, and engineering applicability of sludge-derived fuel formulation design. Summary of the Invention

[0040] In view of this, the present invention provides a machine learning-based method for optimizing sludge-derived briquette fuel formulations. By constructing a sludge-derived briquette fuel dataset, establishing a machine learning prediction model, and combining multi-objective optimization, the method achieves rapid prediction, optimization screening, and feedback correction of sludge-derived briquette fuel formulations, thereby obtaining sludge-derived briquette fuel formulations with high calorific value, good molding strength, stable combustion performance, and low pollutant release.

[0041] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0042] A machine learning-based method for optimizing sludge-derived biomass fuel formulations includes the following steps:

[0043] S1: Collect basic physicochemical parameters of sludge raw materials and auxiliary raw materials, conduct TG-FTIR and TG-MS experiments of combined combustion with different ratios, and obtain data related to combustion characteristic parameters and pollutant generation characteristics.

[0044] S2: Collect data on the physical properties, proportions, combustion characteristics, and flue gas concentration of sludge co-firing based on the Web of Science database. After preprocessing the data, supplement it with the experimental data and combine them to form a dataset of sludge-derived molded fuel combustion characteristics.

[0045] S3: Using raw material characteristic parameters as input variables and target performance indicators as output variables, establish a sludge-derived fuel performance prediction model;

[0046] The machine learning prediction model is trained, validated, and evaluated to obtain the optimal prediction model;

[0047] S4: Couple the optimal prediction model with a multi-objective optimization algorithm to establish a multi-objective optimization model for sludge-derived biomass formulation;

[0048] Based on the output of the multi-objective optimization model, the following parameters are satisfied: calorific value, minimum blending ratio of municipal sludge, combustion stability, and NO. x Recommended formulations for constraining the release of SO2 and HCl pollutants.

[0049] Preferably, in step S1, the basic physicochemical parameters include moisture content, ash content, volatile matter, fixed carbon, lower heating value, higher heating value, elemental analysis of C, H, O, N, and S, particle size distribution, bulk density, and organic matter content.

[0050] Preferably, in step S1, the combustion characteristic parameters include: ignition temperature, burnout temperature, maximum weight loss rate temperature, comprehensive combustion index, combustion stability index, and apparent activation energy; the pollutant formation characteristic parameters include NO in the combustion flue gas. x The amount of SO2, HCl, etc. generated.

[0051] Preferably, in step S2, the preprocessing of the sludge-derived molded fuel combustion characteristic dataset includes: standardization and logarithmic transformation.

[0052] Through the above operations, standardization helps reduce scale differences among feature parameters, while logarithmic transformation helps mitigate the impact of extreme values ​​on the model, thereby improving the accuracy and generalization ability of the prediction model.

[0053] Preferably, in step S3, the raw material characteristic parameters include: industrial analysis, elemental analysis, calorific value, formulation ratio, and heating rate; the target performance indicators include: lower heating value, ignition temperature, burnout temperature, stable combustion index, comprehensive combustion index, and NO. x SO2 and HCl release amounts.

[0054] Preferably, in step S3, the optimal machine learning prediction model is evaluated using the coefficient of determination R², mean absolute error MAE, and root mean square error RMSE.

[0055] Preferably, in step S4, the constraints include: the lower heating value of the molten fuel ≥ 14.6 MJ / kg; and the proportion of municipal sludge ≥ 50%.

[0056] Preferably, in step S4, the optimization of the derivative fuel forming process parameters includes the following steps:

[0057] a: Based on the optimized fuel formulation, a batch molding test of sludge-based briquettes was carried out, including drying, mixing, molding, and drying, to obtain the product's compressive strength, drop strength, abrasion resistance, molding qualification rate, molding moisture content, and molding power consumption performance indicators.

[0058] b: Multi-objective optimization of molding process parameters: Establish a multi-objective optimization model for the molding process, set constraints for multi-objective optimization, and obtain the optimal range of process parameters after optimization.

[0059] Preferably, the molding process parameters include: mixing time, moisture content adjustment, molding pressure, molding temperature, mold hole size, drying temperature, drying time, and real-time statistics of the production power consumption of the molding process.

[0060] The present invention achieves the following technical effects compared to the prior art:

[0061] First, it has excellent adaptability to sludge quality fluctuations. By incorporating sludge moisture content, ash content, calorific value, and elemental composition into the input characteristic system, it can quickly complete the modification of derivative fuel formulas for sludge from different sources and batches, with strong adaptability.

[0062] Second, the efficiency of formula development has been significantly improved. Relying on machine learning models, the performance of molten fuel can be quickly predicted. Coupled with a visual GUI interface to support operation, the workload of repetitive experiments in the traditional R&D model has been greatly reduced, the formula screening cycle has been effectively shortened, and the overall technology has good promotion potential.

[0063] Third, it achieves multi-objective synergistic optimization. By combining the molding process optimization, it can simultaneously take into account multiple dimensions of the derived fuel, such as calorific value, strength, combustion stability, comprehensive combustion characteristics and pollutant release level, thus solving the problem of insufficient engineering adaptability caused by single-index optimization.

[0064] Fourth, it has the ability to dynamically iterate and optimize. Through continuous feedback from experimental verification results and actual operation data, it can continuously update the formula database and prediction model, thereby continuously improving prediction accuracy and engineering adaptability. Attached Figure Description

[0065] Figure 1 A schematic diagram of the process for optimizing sludge-derived fuel formulation based on machine learning provided by the present invention;

[0066] Figure 2 This is a graphical user interface (GUI) diagram of the sludge-derived fuel combustion characteristic prediction program in this invention. Detailed Implementation

[0067] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] This invention discloses a method for optimizing sludge-derived biomass fuel formulations based on machine learning, comprising the following steps:

[0069] S1: Collect basic physicochemical parameters of sludge raw materials and auxiliary raw materials, conduct TG-FTIR and TG-MS experiments of combined combustion with different ratios, and obtain data related to combustion characteristic parameters and pollutant generation characteristics.

[0070] S2: Collect data on the physical properties, proportions, combustion characteristics, and flue gas concentration of sludge co-firing based on the Web of Science database. After preprocessing the data, supplement it with the experimental data and combine them to form a dataset of sludge-derived molded fuel combustion characteristics.

[0071] S3: Using raw material characteristic parameters as input variables and target performance indicators as output variables, establish a sludge-derived fuel performance prediction model;

[0072] The machine learning prediction model is trained, validated, and evaluated to obtain the optimal prediction model;

[0073] S4: Couple the optimal prediction model with a multi-objective optimization algorithm to establish a multi-objective optimization model for sludge-derived biomass formulation;

[0074] Based on the output of the multi-objective optimization model, the following parameters are satisfied: calorific value, minimum blending ratio of municipal sludge, combustion stability, and NO. x Recommended formulations for constraining the release of SO2 and HCl pollutants.

[0075] In step S1, the basic physicochemical parameters include moisture content, ash content, volatile matter, fixed carbon, lower heating value, higher heating value, elemental analysis of C, H, O, N, and S, particle size distribution, bulk density, and organic matter content.

[0076] In step S1, the combustion characteristic parameters include: ignition temperature, burnout temperature, maximum weight loss rate temperature, comprehensive combustion index, combustion stability index, and apparent activation energy; the pollutant formation characteristic parameters include NO in the combustion flue gas. x The amount of SO2, HCl, etc. generated.

[0077] In step S2, the preprocessing of the sludge-derived molded fuel combustion characteristic dataset includes: standardization and logarithmic transformation.

[0078] In step S3, the raw material characteristic parameters include: industrial analysis, elemental analysis, calorific value, formulation ratio, and heating rate; the target performance indicators include: lower heating value, ignition temperature, burnout temperature, stable combustion index, comprehensive combustion index, and NO. x SO2 and HCl release amounts.

[0079] In step S3, the optimal machine learning prediction model is evaluated using the coefficient of determination R², mean absolute error MAE, and root mean square error RMSE.

[0080] In step S4, the constraints include: the lower heating value of the molten fuel ≥ 14.6 MJ / kg; and the proportion of municipal sludge ≥ 50%.

[0081] In step S4, the optimization of the derivative fuel forming process parameters includes the following steps:

[0082] a: Based on the optimized fuel formulation, a batch molding test of sludge-based briquettes was carried out, including drying, mixing, molding, and drying, to obtain the product's compressive strength, drop strength, abrasion resistance, molding qualification rate, molding moisture content, and molding power consumption performance indicators.

[0083] b: Multi-objective optimization of molding process parameters: Establish a multi-objective optimization model for the molding process, set constraints for multi-objective optimization, and obtain the optimal range of process parameters after optimization.

[0084] The molding process parameters include: mixing time, moisture content adjustment, molding pressure, molding temperature, mold hole size, drying temperature, drying time, and real-time statistics of the production power consumption of the molding process.

[0085] Example 1:

[0086] This embodiment uses municipal sludge, industrial sludge, and biomass powder as auxiliary fuels. Based on the physicochemical properties of the raw materials, the compatibility ratio, and thermogravimetric combustion experimental data, a predictive model for the combustion characteristics of sludge-derived fuels is established using the Python programming language. Recommended formulas are obtained through formula traversal and constraint screening.

[0087] One or more types of sludge raw materials and auxiliary fuels are collected, and after drying, crushing and sieving, the sample particle size is controlled to be below 0.15 mm.

[0088] Industrial analysis, elemental analysis, and calorific value testing were conducted on each individual raw material to obtain raw material characteristic parameters such as ash content, volatile matter, fixed carbon, C, H, O, N, S content, and lower heating value.

[0089] Subsequently, with the total mass of sludge raw materials and auxiliary fuel as 100%, sludge mass ratios of 50%, 70%, and 90% were set, and auxiliary fuel mass ratios of 50%, 30%, and 10% were set accordingly to prepare sludge-derived fuel samples.

[0090] Thermogravimetric combustion experiments were conducted on the above-mentioned compatible samples.

[0091] Weigh the sample and place it in an alumina crucible. Use air as the combustion atmosphere, set the air flow rate to 50-100 mL / min, set the heating range to room temperature to 900 ℃, and set the heating rate to one or more of 10, 20, and 30 ℃ / min.

[0092] The TG / DTG curves and pollutant formation characteristics during the combustion process of the samples were obtained by TG-FTIR and TG-MS, and the ignition temperature, burnout temperature, maximum weight loss rate, temperature corresponding to the maximum weight loss rate, average weight loss rate, comprehensive combustion index and combustion stability index were calculated accordingly.

[0093] For samples tested at multiple heating rates, the apparent activation energy was further calculated using the isoconversion method. The formulas for calculating the comprehensive combustion index and combustion stability index are as follows:

[0094] (1)

[0095] (2)

[0096] In the formula, DTG max The maximum rate of weight loss during combustion; DTG mean It is the average rate of weightlessness; T max The temperature at which the maximum rate of weightlessness is achieved; T i and T b These are the ignition temperature and the burnout temperature, respectively.

[0097] The raw material characteristic parameters, compatibility ratios, heating rates, and thermogravimetric combustion results were compiled into a machine learning dataset. This dataset was then supplemented with literature data on sludge co-incineration from databases such as Web of Science. After screening and preprocessing, the dataset was enriched and added to the basic dataset.

[0098] Input variables include ash content, volatile matter, fixed carbon, C, H, O, N, S content, lower heating value, mass ratio of each raw material, and heating rate of sludge raw material and auxiliary fuel;

[0099] The output variables include ignition temperature, burnout temperature, maximum weight loss rate, overall combustion index, and combustion stability index.

[0100] If the TG curve is used as the prediction object, the TG curve is interpolated according to a uniform temperature interval to form a mass residual rate sequence as the model output.

[0101] Python is used for data processing and model training, utilizing the Pandas, NumPy, Scikit-learn, and PyTorch toolkits.

[0102] First, read the experimental data table and remove data with many missing values ​​or incomplete test conditions; for a small number of missing physicochemical property parameters, fill them in with the mean or median of similar raw materials.

[0103] The input variables were then standardized using StandardScaler, and the training, validation, and test sets were divided into sets of 70%, 20%, and 10%, respectively.

[0104] A combustion characteristic prediction model is established using a feedforward neural network based on PyTorch. The neural network includes an input layer, hidden layers, and an output layer. The number of nodes in the input layer is the same as the number of input features. The hidden layer is set to three layers with 128, 64, and 32 nodes respectively. Each hidden layer is followed by a ReLU activation function and a Dropout layer. The number of nodes in the output layer is determined according to the prediction task.

[0105] The model loss function is the mean squared error loss function, the optimizer is the Adam optimizer, the initial learning rate is set to 0.001, the batch size is set to 16 or 32, and the maximum number of training epochs is set to 1000-3000 epochs. During training, the validation set loss is used as the basis for saving the model. Training stops when the validation set loss no longer decreases for several consecutive epochs, and the optimal model parameters are saved.

[0106] After the model training is completed, the prediction effect is evaluated using a test set. The evaluation indicators include the coefficient of determination R², the mean absolute error MAE and the root mean square error RMSE, and their calculation formulas are as shown in (3)-(4).

[0107] For ignition temperature, burnout temperature, maximum weight loss rate, comprehensive combustion index, and combustion stability index, independent prediction models can be established separately, or multi-output prediction models can be established. In this embodiment, an independent single-output model is preferred to reduce the impact of the difference in the dimensions of different output indicators on model training.

[0108] (3)

[0109] (4)

[0110] After obtaining the optimal prediction model, a candidate formulation space is set, in which the mass ratio of sludge is not less than 50%, the mass ratio of auxiliary fuel is not more than 50%, and the sum of the mass ratios of each raw material is 100%.

[0111] Candidate formulations are generated in the candidate formulation space with a step size of 1% or 5%, and the corresponding input features are calculated based on the measured characteristic parameters of each raw material. These features are then input into the trained prediction model to obtain the predicted results of ignition temperature, burnout temperature, comprehensive combustion index, combustion stability index, and lower heating value of the candidate formulations.

[0112] Candidate formulations were screened under constraints. Screening criteria included: a lower heating value not lower than a set threshold, preferably not lower than 14.6 MJ / kg; and comprehensive combustion index and combustion stability index meeting set requirements.

[0113] For candidate formulations that meet the above conditions, those with higher overall flammability index, better combustion stability, and higher sludge blending ratio are preferred as recommended formulations; when NO is introduced... x When collecting data on SO2 or HCl release, the pollutant release amount should not exceed the set limit as a screening criterion.

[0114] For the recommended formulation, molding verification experiments were conducted. Sludge raw materials and auxiliary fuel were mixed according to the recommended ratio, with limestone, bentonite, starch, or other desulfurizing agents, deacidifying agents, and binders added if necessary. After moisture conditioning, the mixture was pressed into molds at a pressure of 5-20 MPa and a molding temperature of room temperature to 120°C, followed by drying at 60-105°C to the target moisture content. The compressive strength, drop strength, abrasion resistance, molding pass rate, and molding moisture content of the obtained sludge-derived briquettes were tested.

[0115] If the molded fuel meets the strength and moisture content requirements, the formula is determined as the recommended formula; if it does not meet the requirements, the binder ratio, moisture content, molding pressure or drying conditions are adjusted, and the new experimental results are added to the database.

[0116] After each batch of recommended formulations has completed thermogravimetric analysis (TGA) or molding verification, the raw material properties, compatibility ratios, TGA characteristics, and molding performance of the new samples are added to the original database. Data standardization, model training, and model evaluation are then performed again to ensure that the prediction model is continuously updated as experimental data increases, thereby improving its adaptability to sludge and auxiliary fuels from different sources.

[0117] In this embodiment, the model training results show that the coefficient of determination (R²) for each output index on the test set is above 0.9, and the prediction error meets engineering requirements, indicating that the model has good prediction accuracy and can accurately reflect the nonlinear mapping relationship between raw material characteristics, formulation ratio, and combustion performance. Finally, an optimized formulation that meets the calorific value constraint and has a sludge blending ratio of 70% was selected. Its predicted comprehensive combustion index meets the design requirements, and all physical properties of the briquetted fuel meet the application standards for municipal sludge-derived briquetted fuel. The deviation between the verification results and the model prediction values ​​is within the allowable range, proving the reliability of this method.

[0118] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the technical scope of the present invention. Therefore, any minor modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for optimizing sludge-derived biomass fuel formulation based on machine learning, characterized in that, Includes the following steps: S1: Collect basic physicochemical parameters of sludge raw materials and auxiliary raw materials, conduct TG-FTIR and TG-MS experiments of combined combustion with different ratios, and obtain data related to combustion characteristic parameters and pollutant generation characteristics. S2: Collect data on the physical properties, proportions, combustion characteristics, and flue gas concentration of sludge co-firing based on the Web of Science database. After preprocessing the data, supplement it with experimental data and combine them to form a dataset of sludge-derived molded fuel combustion characteristics. S3: Using raw material characteristic parameters as input variables and target performance indicators as output variables, establish a sludge-derived fuel performance prediction model; The machine learning prediction model is trained, validated, and evaluated to obtain the optimal prediction model; S4: Couple the optimal prediction model with a multi-objective optimization algorithm to establish a multi-objective optimization model for sludge-derived biomass formulation; Based on the output of the multi-objective optimization model, the following parameters are satisfied: calorific value, minimum blending ratio of municipal sludge, combustion stability, and NO. x Recommended formulations for constraining the release of SO2 and HCl pollutants.

2. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S1, the basic physicochemical parameters include moisture content, ash content, volatile matter, fixed carbon, lower heating value, higher heating value, elemental analysis of C, H, O, N, and S, particle size distribution, bulk density, and organic matter content.

3. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S1, the combustion characteristic parameters include: ignition temperature, burnout temperature, maximum weight loss rate temperature, comprehensive combustion index, combustion stability index, and apparent activation energy; the pollutant formation characteristic parameters include NO in the combustion flue gas. x The amount of SO2, HCl, etc. generated.

4. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S2, the preprocessing of the sludge-derived molded fuel combustion characteristic dataset includes: standardization and logarithmic transformation.

5. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S3, the raw material characteristic parameters include: industrial analysis, elemental analysis, calorific value, formulation ratio, and heating rate; the target performance indicators include: lower heating value, ignition temperature, burnout temperature, stable combustion index, comprehensive combustion index, and NO. x SO2 and HCl release amounts.

6. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S3, the optimal machine learning prediction model is evaluated using the coefficient of determination R², mean absolute error MAE, and root mean square error RMSE.

7. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S4, the constraints include: the lower heating value of the molten fuel ≥ 14.6 MJ / kg; and the proportion of municipal sludge ≥ 50%.

8. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 1, characterized in that, In step S4, the optimization of the derivative fuel forming process parameters includes the following steps: a: Based on the optimized fuel formulation, a batch molding test of sludge-based briquettes was carried out, including drying, mixing, molding and drying, to obtain the product's compressive strength, drop strength, abrasion resistance, molding qualification rate, molding moisture content and molding power consumption performance indicators. b: Multi-objective optimization of molding process parameters: Establish a multi-objective optimization model for the molding process, set constraints for multi-objective optimization, and obtain the optimal range of process parameters after optimization.

9. The method for optimizing sludge-derived biomass fuel formulation based on machine learning according to claim 8, characterized in that, The molding process parameters include: mixing time, moisture content adjustment, molding pressure, molding temperature, mold hole size, drying temperature, drying time, and real-time statistics of the production power consumption of the molding process.

Images