Method for converting sewage into biofuel for power generation

Abstract

The invention relates to the technical field of sewage treatment and biological energy, and particularly discloses a method for converting sewage into biofuel for power generation. According to the method, a multi-factor synergistic system containing an inducing factor, a phagocytosis factor and a powdered sugar factor is generated through photosynthetic induction, the inducing factor destroys harmful microorganisms, the phagocytosis factor fixes radioactive substances, germs and viruses, and the powdered sugar factor serves as a carbon value locking precursor; organic pollutants are directionally converted into alcohol clean energy carriers by utilizing a multi-factor synergistic system, and meanwhile, fixed capture of the pollutants is completed; the carbon value is locked by a carbon value locking agent to generate a fixed carbon mixture, and the fixed carbon mixture is dehydrated and dried to obtain the high-purity biochar fuel. Unification of harmless treatment and energy utilization of the radioactive sewage is realized, and the limitation that the radioactive sewage cannot be treated in the traditional technology is broken through.

Description

Technical Field

[0001] This invention relates to the fields of wastewater treatment and bioenergy technology, and in particular to a method for converting wastewater into biofuel for power generation. Background Technology

[0002] In the field of wastewater treatment and energy recovery, converting organic matter in wastewater into biofuels is an important way to achieve resource utilization. Existing technologies mainly employ anaerobic digestion to produce methane, microbial fuel cells to generate electricity, or pyrolysis to produce bio-oil. Anaerobic digestion technology converts organic matter into methane through microbial metabolism and has been widely used in the treatment of high-concentration organic wastewater. Microbial fuel cell technology utilizes electrogenic bacteria to directly degrade organic matter and generate electricity. Pyrolysis technology decomposes organic matter in wastewater into bio-oil and biochar at high temperatures. These technologies have achieved certain results in the treatment of conventional municipal wastewater and industrial organic wastewater.

[0003] However, existing technologies all face a fundamental flaw: they cannot simultaneously achieve energy utilization while harmlessly treating radioactive pollutants. Radioactive wastewater contains both organic pollutants and radionuclides. Anaerobic digestive microorganisms are inactivated by radiation, leading to system collapse. The electrogenic bacteria in microbial fuel cells are also sensitive to radiation, and the pyrolysis process cannot fix radionuclides; instead, it may cause them to escape with the flue gas, resulting in secondary pollution. Furthermore, existing technologies produce gaseous methane or liquid bio-oil, which suffers from high storage and transportation costs, poor combustion stability, and easy carbon escape, failing to achieve permanent carbon fixation. This invention addresses these fundamental flaws by proposing a novel method that uses photosynthetic induction to generate a multi-factor synergistic system. This system simultaneously completes the fixation of radioactive pollutants and the directional conversion of organic pollutants into solid biocarbon within the same system, achieving a technological paradigm leap from "wastewater treatment" to "carbon sequestration."

[0004] Therefore, this invention proposes a method for converting wastewater into biofuel for power generation. Summary of the Invention

[0005] This invention provides a method for converting wastewater into biofuel for power generation. By using a photosynthetic-induced multi-factor synergistic system, it achieves the unification of harmless treatment of radioactive pollutants and directional conversion of organic pollutants into highly stable solid biocarbon fuel. This method overcomes the limitations of traditional technologies that cannot treat radioactive wastewater and produces high-purity biocarbon fuel that can be directly used for power generation.

[0006] This invention provides a method for converting wastewater into biofuel for power generation, comprising the following steps: Wastewater is introduced into a photosynthetic reactor to enrich carbon sources. Under artificially enhanced photosynthesis, organic carbon, inorganic carbon and trace elements in the wastewater are mixed to form a highly active mixed carbon system. In a highly active mixed carbon system, a multi-factor synergistic system containing inducing factors, phagocytic factors, and sugar powder factors is generated through catalytic reaction. The inducing factors are used to destroy the cell structure of harmful microorganisms, the phagocytic factors are used to fix radioactive materials and pathogens, and the sugar powder factors are used as carbon value locking precursors. A multi-factor synergistic system is used to harmlessly transform wastewater, converting organic pollutants into clean energy carriers of alcohols, while simultaneously immobilizing and capturing radioactive substances and pathogens. After the harmless transformation is completed in the multi-factor synergistic system, the carbon value is locked by a carbon value locking agent to generate a fixed carbon mixture containing phagocytic factor solidified body and sugar powder factor polymer. A mixture of stationary carbons is dehydrated and dried to obtain high-purity biocarbon fuel.

[0007] Furthermore, the carbon source enrichment step includes constructing a closed loop for the dynamic regulation of photosynthetic active bacterial communities: The photosynthetic active bacteria are inoculated into the photosynthetic reaction tank. The photosynthetic active bacteria are composed of Chromobacteriaceae, Rhodospirilluae, and Chlorophyticusaceae in a preset ratio. The preset ratio is dynamically determined by a carbon source photosyntheticity prediction model based on the carbon source composition in the wastewater. During the carbon source enrichment process, the number of viable bacteria and the distribution of community structure of photosynthetic active bacteria were monitored in real time by online flow cytometry to obtain real-time monitoring data of bacterial community activity. Real-time monitoring data of microbial community activity is input into the photosynthetic active microbial community metabolic dynamics sub-model, which is constructed based on the principle of microbial quorum sensing and includes state variables of electron transfer flux among microbial communities. The deviation between the current metabolic activity of the photosynthetic microbial community and the expected metabolic pathway is analyzed in real time using a sub-model of the metabolic dynamics of the photosynthetic microbial community. When the deviation exceeds the first preset threshold, the pre-stored control strategy library is matched according to the deviation feature vector to generate and execute dynamic control instructions including light intensity adjustment, trace element supplementation, and microbial community supplementation ratio.

[0008] Furthermore, the carbon source photosynthetic capacity prediction model is constructed and dynamically updated in the following manner: We collected a large amount of historical batches of wastewater using three-dimensional fluorescence spectroscopy-high performance liquid chromatography-mass spectrometry full-spectrum data, and extracted fluorescence component scores representing different carbon source components as carbon source feature vectors through parallel factor analysis. The carbon source enrichment rate and the maximum specific growth rate of photosynthetic bacteria under standard photosynthetic induction conditions were collected from each historical batch of wastewater as carbon source photosyntheticity labels. Using carbon source feature vectors as input and carbon source photosyntheticity labels as output, a gradient boosting regression tree model is trained to obtain an initial carbon source photosyntheticity prediction model. After each batch of wastewater is treated, the carbon source feature vector and the measured photosynthetic carbon source data of that batch of wastewater are used as new samples to incrementally learn and update the gradient boosting regression tree model.

[0009] Furthermore, the steps for generating a multi-factor synergistic system include constructing a closed-loop sub-loop of induced factor generation dynamics: A composite photocatalyst was added to a highly active mixed carbon system. The composite photocatalyst was a nano-titanium dioxide supported on iron ions and nitrogen elements and was doped and modified. The amount added was determined by the concentration of dissolved organic carbon in the mixed carbon system through an induction factor demand model. The catalytic reaction was carried out under combined ultraviolet and visible light irradiation conditions, and the intensity sequence of characteristic fluorescence peaks of the inducing factor was monitored in real time by an online three-dimensional fluorescence spectrometer. The characteristic fluorescence peak intensity sequence is input into the inducible factor generation kinetics sub-model, which is constructed based on the photocatalytic oxidation kinetics theory and includes the state variables of hydroxyl radical generation rate and consumption rate. The deviation between the current generation rate and the theoretical maximum generation rate of the induced factor is analyzed in real time using a sub-model of the kinetics of induced factor generation. When the rate deviation exceeds the second preset threshold, the light intensity adjustment is calculated and executed by the proportional-integral-derivative controller based on the rate deviation value, so that the generation rate of the inducing factor is maintained within the preset target range.

[0010] Furthermore, the harmless transformation steps include constructing a closed loop of dynamic adsorbents for phagocytic factors: During the multi-factor synergistic process, the total radioactivity intensity decay curve of the reaction system is monitored in real time using an online radioactivity detector; The decay curves of viable bacteria and viruses in the reaction system were monitored in real time by combining online flow cytometry with plaque counting. The radioactivity intensity decay curve and the viable bacteria count decay curve were input into the phagocytic factor adsorption kinetics sub-model. The phagocytic factor adsorption kinetics sub-model was constructed based on the Langmuir adsorption isotherm equation and the pseudo-second-order adsorption kinetics equation and included the occupancy rate of the phagocytic factor surface active sites as a state variable. The adsorption efficiency coefficient between the current adsorption rate and the theoretical maximum adsorption rate is analyzed in real time using a phagocytic factor adsorption kinetics sub-model. When the adsorption efficiency coefficient is lower than the third preset threshold, the phagocytic factor replenishment instruction is generated and executed based on the adsorption efficiency coefficient through the fuzzy inference rule base. The input variables of the fuzzy inference rule base are the adsorption efficiency coefficient and the ratio of the current radioactivity intensity to the initial radioactivity intensity.

[0011] Furthermore, the carbon value locking step includes constructing a closed-loop sub-loop that regulates the degree of polymerization of the sugar powder factor: During the curing reaction with the addition of a carbon number locking agent, the particle size distribution sequence of the sugar powder factor polymer was monitored in real time using an online dynamic light scattering instrument; The intensity variation sequence of absorption peaks of characteristic functional groups in the sugar powder factor polymer was monitored in real time using an online Fourier transform infrared spectrometer. The characteristic functional groups include hydroxyl, carbonyl, and ether bonds. The particle size distribution sequence and the intensity variation sequence of the absorption peak of the characteristic functional group are input into the polymerization kinetics sub-model of the sugar powder factor. The polymerization kinetics sub-model of the sugar powder factor is constructed based on the condensation reaction mechanism and includes the state variable of the degree of polymerization distribution. The deviation between the current polymerization progress and the theoretical complete polymerization progress is analyzed in real time using a sugar powder factor polymerization kinetics sub-model. When the progress deviation exceeds the fourth preset threshold, the combined optimization command of the carbon number locking agent replenishment acceleration adjustment amount, stirring speed adjustment amount, and reaction temperature adjustment amount is calculated and executed based on the progress deviation value through the model prediction control algorithm.

[0012] Furthermore, the generation of inducible factors also includes steps for maintaining activity and regulating directional effects: The steady-state concentration of hydroxyl radicals in the inducing factor was monitored in real time using online chemiluminescence. The steady-state concentration of hydroxyl radicals is compared with a preset activity threshold to generate an activity deviation value; When the activity deviation value exceeds the fifth preset threshold, the dissolved oxygen replenishment amount or the co-catalytic aid dosage is calculated by the fuzzy logic controller based on the activity deviation value. The co-catalytic aid is persulfate or hydrogen peroxide. Execute control commands based on the calculated dissolved oxygen replenishment or co-catalytic agent dosage to maintain the oxidative destructive activity of inducing factors on the cell structure of harmful microorganisms. Real-time monitoring of cell morphology changes in harmful microorganisms using an online particle imager yields time-series data on cell damage rate. The time-series data of cell damage rate is compared with the preset target damage rate. When the cell damage rate is lower than the sixth preset threshold, the irradiation intensity of the photocatalytic reaction is automatically increased or the photocatalytic reaction time is extended.

[0013] Furthermore, the dehydration and drying process includes a biocarbon fuel quality grading and reverse traceability sub-loop: High-purity biocarbon fuel was tested for fixed carbon content, calorific value, ash content, volatile matter, and heavy metal leaching toxicity to obtain multidimensional quality parameters of the biocarbon fuel. The multidimensional quality parameters of biocarbon fuel are compared with the preset target quality parameters to generate a quality deviation vector; The quality deviation vector is input into the quality deviation tracing model based on deep belief network. The hidden nodes of the deep belief network correspond to the key control parameters of each process step. The contribution distribution of each process step to the quality deviation is obtained by backpropagation of the network. Based on the contribution distribution, the key process steps and their key control parameters that lead to quality deviations are identified, and traceability results are generated. The traceability results are fed back to the corresponding process step's sub-closed-loop control system, serving as a reference weight factor for subsequent control of that sub-closed-loop control system.

[0014] Furthermore, the quality deviation tracing model is constructed and updated in the following ways: Collect a large number of historical batches of wastewater treatment process parameters and corresponding high-purity biocarbon fuel multidimensional quality parameters to build a historical case library; For each case in the historical case library, label the actual cause of the quality deviation. The actual cause labels include insufficient carbon source enrichment, insufficient generation of inducing factors, saturation of phagocytic factors, incomplete polymerization of sugar powder factors, and excessively high dehydration and drying temperature. Using the sequence of process parameters throughout the entire process as input and the actual cause labels as output, a gradient boosting decision tree classifier is trained to obtain an initial quality deviation tracing model. After each batch of wastewater treatment is completed, the entire process parameter sequence of that batch and the actual cause labels after manual verification are used as new samples to incrementally learn and update the gradient boosting decision tree classifier.

[0015] Furthermore, it also includes a top-level closed loop for end-to-end collaborative optimization based on reinforcement learning: The dynamic regulation sub-loop of photosynthetic active bacteria community, the kinetics sub-loop of inducing factor generation, the regulation sub-loop of inducing factor activity maintenance, the dynamic adsorption sub-loop of phagocytic factor, the polymerization degree regulation sub-loop of sugar powder factor, and the quality grading and reverse traceability sub-loop of biocarbon fuel are jointly constructed into a multi-agent collaborative optimization system. Each sub-loop corresponds to an agent. The action space of each agent is the range of values ​​for its control commands, and the state space of each agent is the feature vector of its real-time monitoring data. Construct a central coordinating agent. The input of the central coordinating agent is the state vector of each sub-closed-loop agent and the initial water quality parameters of the current batch of wastewater. The output of the central coordinating agent is the collaborative weight coefficient of each sub-closed-loop agent. Design a global reward function, which is a weighted sum of the biocarbon fuel yield reward, biocarbon fuel quality reward, unit energy consumption penalty, and processing time penalty. Each sub-closed-loop agent adjusts its own policy network objective function according to the collaborative weight coefficients allocated by the central coordinating agent, and performs collaborative training through a multi-agent deep deterministic policy gradient algorithm to maximize the global reward function value. The multi-agent collaborative optimization system, after training convergence, is deployed in the actual wastewater treatment process to achieve online collaborative optimization and control of the entire process.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: This invention utilizes a multi-factor synergistic system, including inducing factors, phagocytic factors, and glycogen factors, to simultaneously achieve radionuclide fixation, pathogen and virus inactivation, and the directional conversion of organic pollutants into solid biocarbon. This allows for the unified realization of harmless treatment and energy utilization of radioactive wastewater within the same system. The high-purity biocarbon fuel produced by this method can be stored stably for a long time at normal temperature and pressure, eliminating the need for high-pressure storage and transportation equipment. It exhibits stable combustion performance and can be directly used for biomass power generation. Simultaneously, the multi-factor synergistic system achieves energy self-sufficiency, significantly reducing treatment energy consumption. The solidified radioactive materials meet national leaching toxicity standards, eliminating the risk of secondary pollution and realizing a fundamental shift in wastewater treatment from carbon cycling to carbon sequestration.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[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] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a core process flow diagram of the method for converting wastewater into biofuel for power generation in an embodiment of the present invention; Figure 2 This is an integrated architecture diagram of the six sub-loops and the main process in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] refer to Figure 1 and Figure 2 This invention provides a method for converting wastewater into biofuel for power generation, comprising the following steps: Wastewater is introduced into a photosynthetic reactor to enrich carbon sources. Under artificially enhanced photosynthesis, organic carbon, inorganic carbon and trace elements in the wastewater are mixed to form a highly active mixed carbon system. In a highly active mixed carbon system, a multi-factor synergistic system containing inducing factors, phagocytic factors, and sugar powder factors is generated through catalytic reaction. The inducing factors are used to destroy the cell structure of harmful microorganisms, the phagocytic factors are used to fix radioactive materials and pathogens, and the sugar powder factors are used as carbon value locking precursors. A multi-factor synergistic system is used to harmlessly transform wastewater, converting organic pollutants into clean energy carriers of alcohols, while simultaneously immobilizing and capturing radioactive substances and pathogens. After the harmless transformation is completed in the multi-factor synergistic system, the carbon value is locked by a carbon value locking agent to generate a fixed carbon mixture containing phagocytic factor solidified body and sugar powder factor polymer. A mixture of stationary carbons is dehydrated and dried to obtain high-purity biocarbon fuel.

[0022] In this embodiment, introducing wastewater into a photosynthetic reactor for carbon source enrichment refers to the process of artificially enhancing photosynthesis to fully mix organic carbon, inorganic carbon, and trace elements such as iron, manganese, and zinc in the wastewater, forming a highly bioactive mixed carbon system. The photosynthetic reactor is pre-inoculated with domesticated photosynthetic bacteria, a mixture of *Rhodopseudomonas palustris*, *Rhodospirillum rubrum*, and *Sulphurella multocida* in a predetermined ratio. This predetermined ratio is dynamically determined based on the carbon source composition in the wastewater using a carbon source photosyntheticability prediction model. The carbon source photosyntheticability prediction model is constructed using a gradient boosting regression tree algorithm. The model input is a carbon source feature vector acquired through three-dimensional fluorescence spectroscopy-high performance liquid chromatography-mass spectrometry and extracted via parallel factor analysis. This carbon source feature vector includes scores for tyrosine-like components, tryptophan-like components, fulvic acid-like components, and humic acid-like components. The model output is the carbon source enrichment rate and the maximum specific growth rate of the photosynthetic bacteria. The training parameters for the gradient boosting regression tree model were set as follows: learning rate 0.1, maximum tree depth 5, number of subtrees 100, and minimum number of leaf samples 5. Five-fold cross-validation was used to prevent overfitting, and a coefficient of determination greater than 0.85 was used as the model training pass criterion. After each batch of wastewater was treated, the carbon source feature vector, measured carbon source enrichment rate, and maximum specific growth rate of photosynthetic active bacteria in that batch of wastewater were used as new samples to incrementally update the gradient boosting regression tree model. The weight of the new samples was set to 1.0, and the weight of the historical samples was decreased successively with a decay coefficient of 0.95.

[0023] In this embodiment, the generation of a multi-factor synergistic system containing inducing factors, phagocytic factors, and sugar powder factors through a catalytic reaction in a highly active mixed carbon system refers to the addition of a doped and modified catalyst, nano-titanium dioxide loaded with iron ions and nitrogen elements, to the mixed carbon system. The iron ion doping amount accounts for 0.5% to 2% of the mass of titanium dioxide, and the nitrogen element doping amount accounts for 1% to 3% of the mass of titanium dioxide. The photocatalytic reaction occurs under combined ultraviolet and visible light irradiation conditions, with ultraviolet light wavelengths of 254 nm to 365 nm, visible light wavelengths of 400 nm to 700 nm, an ultraviolet to visible light intensity ratio of 1:3 to 1:5, and a total irradiation intensity of 200 W / m² to 500 W / m². The generated hydroxyl radicals act as inducing factors to oxidatively destroy the cell membrane structure of harmful microorganisms. Simultaneously, chitosan-modified materials, quaternary ammonium salt functionalized polymers, and zeolite molecular sieves in the mixed carbon system are compounded at mass percentages of 30% to 50%, 20% to 40%, and 20% to 40% to form phagocytic factors. The chitosan-modified material is carboxymethyl chitosan with a degree of substitution of 0.8 to 1.2 and a molecular weight of 100,000 to 300,000, which undergoes coordination chelation reactions with radioactive nuclides through amino groups. The quaternary ammonium salt functionalized polymer is polydimethyldiallylammonium chloride with a molecular weight of 500,000 to 1,000,000, which electrostatically adsorbs the negatively charged viral capsid through quaternary ammonium cations and inserts into the bacterial cell membrane, causing membrane rupture. The zeolite molecular sieve is a 13X type molecular sieve with a silica-to-alumina ratio of 2.5 to 3.0 and a pore size of 0.8 nm to 1.0 nm, which fixes radioactive nuclides such as cesium ions and strontium ions through ion exchange. Under acidic conditions, sugars such as glucose, fructose, and sucrose in the mixed carbon system act as sugar powder factors, becoming precursors for subsequent carbon number locking.

[0024] In this embodiment, the harmless transformation of wastewater using a multi-factor synergistic system refers to the continuous attack and lysis of harmful microbial cells by inducing factors, and the fixation of radionuclides and pathogens / viruses on their surfaces or within their pores by chelation, ion exchange, and electrostatic adsorption by phagocytic factors. Simultaneously, organic pollutants are directionally transformed into clean energy carriers such as methanol and ethanol under the action of reactive oxygen species such as hydroxyl radicals and superoxide radicals. During the multi-factor synergistic system operation, the total radioactivity decay curve of the reaction system is monitored in real time using an online radioactivity detector with a detection limit of 0.1 Bq / L and a sampling frequency of once per minute. The viable count decay curve of pathogens and viruses in the reaction system is monitored in real time using an online flow cytometer coupled with plaque counting. The flow cytometer detects 10,000 events per minute, and the plaque counting detection limit is 10 plaque-forming units per milliliter. The radioactivity decay curve and viable count decay curve are input into the phagocytic factor adsorption kinetics sub-model. The phagocytic factor adsorption kinetics sub-model is constructed based on the Langmuir adsorption isotherm equation and the pseudo-second-order adsorption kinetics equation. The nonlinear least squares method is used to fit the model parameters. The model input is the radioactivity intensity decay sequence and the viable bacteria count decay sequence. The model output is the adsorption efficiency coefficient between the current adsorption rate and the theoretical maximum adsorption rate. The adsorption efficiency coefficient ranges from 0 to 1. When the adsorption efficiency coefficient is lower than 0.6, the phagocytic factor supplementation regulation is triggered.

[0025] In this embodiment, after the multi-factor synergistic system completes the harmless transformation, the carbon value is locked by a carbon value locking agent. This refers to adding one or more of the following as carbon value locking agents to the reaction system: phenolic resin prepolymer, lignin sulfonate, and humate. The phenolic resin prepolymer is a thermoplastic phenolic resin with a molecular weight of 500 to 2000 and a solid content of 60% to 70%. The lignin sulfonate is sodium lignin sulfonate with a sulfonation degree of 1.5 to 2.0 mmol / g and a molecular weight of 10,000 to 50,000. The humate is sodium humate with a humic acid content greater than 60% and a molecular weight of 5000 to 20,000. The reaction proceeds under the conditions of pH 3 to 5, temperature 80 to 100 degrees Celsius, and oxalic acid catalyst dosage of 1% to 3% of the sugar powder factor mass, undergoing a condensation reaction with the sugar powder factor. The hydroxyl and carboxyl groups of sugar molecules undergo esterification condensation to remove water molecules and form ester bonds. Sugar molecules are then linked by glycosidic bonds to form oligosaccharide and polysaccharide chains. The hydroxymethyl groups in the phenolic resin prepolymer undergo etherification and cross-linking with the hydroxyl groups of the sugars to form a three-dimensional network structure, generating a sugar powder factor polymer with a network structure. Simultaneously, a solidified phagocytic factor containing immobilized radioactive material is encapsulated within this polymer, forming a structurally stable immobilized carbon mixture. During the polycondensation reaction, the polymer particle size is monitored in real-time using an online dynamic light scattering instrument, with a target particle size range of 1 to 10 micrometers. The intensity of the characteristic ester bond peak at 1730 wavenumber is monitored in real-time using an online Fourier transform infrared spectroscopy instrument. The reaction is considered complete when the peak intensity stabilizes. The typical reaction time is 60 to 120 minutes.

[0026] In this embodiment, obtaining high-purity biocarbon fuel by dehydrating and drying a fixed carbon mixture involves dehydrating the fixed carbon precipitate to a moisture content below 30% using a plate and frame filter press. The filter press pressure is 0.5 MPa to 0.8 MPa, and the filtration cycle is 30 to 60 minutes. The dehydrated filter cake is then conveyed to a fluidized bed dryer and dried at 105 to 120 degrees Celsius to a moisture content below 5%. The dryer inlet air temperature is 120 to 150 degrees Celsius, the outlet air temperature is 60 to 80 degrees Celsius, and the drying time is 20 to 40 minutes, resulting in a high-purity biocarbon fuel dry-based product that can be directly used for biomass power generation. High-purity biocarbon fuel was tested for fixed carbon content, calorific value, ash content, volatile matter, and heavy metal leaching toxicity. Fixed carbon content testing was performed in accordance with GB / T 28731 standard, calorific value testing in accordance with GB / T 30727 standard, ash content testing in accordance with GB / T 30726 standard, volatile matter testing in accordance with GB / T 30730 standard, and heavy metal leaching toxicity testing in accordance with HJ 557 standard. When the fixed carbon content is greater than 85% and the calorific value is greater than 25 MJ / kg, the biocarbon fuel is directly graded as Grade I and transported to the biomass power generation boiler. When the fixed carbon content is between 70% and 85% and the calorific value is between 20 and 25 MJ / kg, the biocarbon fuel is graded as Grade II and blended with Grade I at a mass ratio of 1:3 to 1:1 before being transported to the biomass power generation boiler. When the fixed carbon content is less than 70% or the calorific value is less than 20 MJ / kg, the biocarbon fuel is graded as Grade III and returned to the photosynthetic reactor as a carbon source for secondary conversion. The solidified phagocytic factor meets the requirements of GB 14569.1-2011 standard, with a compressive strength greater than 7 MPa, a leaching rate less than 1 x 10^-3 cm / day, and a free liquid volume less than 0.5%. After passing the test, it is placed in a 200-liter stainless steel bucket and fixed with cement mortar. The cement mortar ratio is 1:2:0.5 (cement to sand to water), and cured for 7 days. The packaging is temporarily stored in a dedicated storage facility with a radiation dose rate of less than 2.5 microsieverts per hour, and is periodically collected by a qualified unit and transported to a disposal site for final disposal.

[0027] Furthermore, the carbon source enrichment step includes constructing a closed loop for the dynamic regulation of photosynthetic active bacterial communities: The photosynthetic active bacteria are inoculated into the photosynthetic reaction tank. The photosynthetic active bacteria are composed of Chromobacteriaceae, Rhodospirilluae, and Chlorophyticusaceae in a preset ratio. The preset ratio is dynamically determined by a carbon source photosyntheticity prediction model based on the carbon source composition in the wastewater. During the carbon source enrichment process, the number of viable bacteria and the distribution of community structure of photosynthetic active bacteria were monitored in real time by online flow cytometry to obtain real-time monitoring data of bacterial community activity. Real-time monitoring data of microbial community activity is input into the photosynthetic active microbial community metabolic dynamics sub-model, which is constructed based on the principle of microbial quorum sensing and includes state variables of electron transfer flux among microbial communities. The deviation between the current metabolic activity of the photosynthetic microbial community and the expected metabolic pathway is analyzed in real time using a sub-model of the metabolic dynamics of the photosynthetic microbial community. When the deviation exceeds the first preset threshold, the pre-stored control strategy library is matched according to the deviation feature vector to generate and execute dynamic control instructions including light intensity adjustment, trace element supplementation, and microbial community supplementation ratio.

[0028] In this embodiment, the carbon source enrichment step includes constructing a closed-loop dynamic regulation sub-loop for photosynthetic active bacterial communities. The photosynthetic reactor is inoculated with acclimatized *Rhodopseudomonas palustris*, *Rhodospirillum rubrum*, and *Sulphurella morganii*. The acclimatization process involves gradually increasing the wastewater concentration, starting at 10% and increasing daily by 10% to 100%, for a total acclimatization period of 10 days. The bacterial community composition ratio is dynamically determined based on the carbon source composition in the wastewater using a carbon source photosyntheticity prediction model. When the volatile fatty acid content is greater than 60%, the proportion of *Rhodopseudomonas palustris* increases to 50% to 60%; when the sugar content is greater than 50%, the proportion of *Rhodospirillum rubrum* increases to 40% to 50%; and when the sulfide content is greater than 50 mg / L, the proportion of *Sulphurella morganii* increases to 30% to 40%.

[0029] In this embodiment, the number of viable bacteria and the distribution of community structure are monitored in real time every 30 minutes using an online flow cytometer. SYBR Green I staining combined with propidium iodide double staining is used to distinguish between viable and dead bacteria, obtaining real-time monitoring data on bacterial activity. The monitoring data is input into a photosynthetic bacterial community metabolic kinetics sub-model constructed based on the principle of microbial quorum sensing. The model uses an extended Kalman filter algorithm to update state variables in real time. The inputs are the viable bacterial concentration sequence and influent water quality parameters, and the outputs are the specific growth rate, substrate consumption rate, electron transport flux, and expected metabolic pathway reference values ​​for each bacterial community.

[0030] In this embodiment, the deviation between the current bacterial community metabolic activity and the expected metabolic pathway is analyzed in real time by a model. The deviation is calculated using a weighted Euclidean distance algorithm, with a value ranging from 0 to 100. A first preset threshold is set to 15 to 25. When the deviation exceeds the threshold, a pre-stored control strategy library is matched based on the deviation feature vector. The strategy library is constructed using a case-based reasoning method, storing over 500 historical case scenarios. The top three cases with the highest similarity are selected using the K-nearest neighbor algorithm and weighted fusion is performed to generate dynamic control instructions, including light intensity adjustment of 500 to 3000 lux, trace element supplementation of 0.1 to 0.5 liters per cubic meter of wastewater, and bacterial community supplementation ratio of 5% to 15%, which are then executed. If the deviation does not recover within 30 minutes after the control instructions are executed, the matching is repeated until normalcy is achieved.

[0031] Furthermore, the carbon source photosynthetic capacity prediction model is constructed and dynamically updated in the following manner: We collected a large amount of historical batches of wastewater using three-dimensional fluorescence spectroscopy-high performance liquid chromatography-mass spectrometry full-spectrum data, and extracted fluorescence component scores representing different carbon source components as carbon source feature vectors through parallel factor analysis. The carbon source enrichment rate and the maximum specific growth rate of photosynthetic bacteria under standard photosynthetic induction conditions were collected from each historical batch of wastewater as carbon source photosyntheticity labels. Using carbon source feature vectors as input and carbon source photosyntheticity labels as output, a gradient boosting regression tree model is trained to obtain an initial carbon source photosyntheticity prediction model. After each batch of wastewater is treated, the carbon source feature vector and the measured photosynthetic carbon source data of that batch of wastewater are used as new samples to incrementally learn and update the gradient boosting regression tree model.

[0032] In this embodiment, the carbon source photosyntheticity prediction model is constructed using a gradient boosting regression tree algorithm. Three-dimensional fluorescence spectroscopy-high performance liquid chromatography-mass spectrometry (HPLC-MS) full-spectrum data of historical batches of wastewater were collected. The three-dimensional fluorescence spectroscopy acquisition conditions were excitation wavelength 200–450 nm and emission wavelength 250–550 nm. HPLC used a C18 column and a UV detector wavelength of 254 nm. The mass spectrometry scan range was a mass-to-charge ratio of 50–1000. Parallel factor analysis was used to extract scores for four fluorescent components—tyrosine-like, tryptophan-like, fulvic acid-like, and humic acid-like—as a 4-dimensional carbon source feature vector. The parallel factor analysis employed alternating least squares, and the number of factors was determined when the kernel consistency was greater than 90%.

[0033] In this embodiment, the carbon source enrichment rate and the maximum specific growth rate of photosynthetically active bacteria in each historical batch of wastewater under standard photosynthetic induction conditions were collected as photosynthetic capability labels. Standard conditions were set as follows: light intensity 10,000 lux, photoperiod-to-dark ratio 12 hours to 12 hours, temperature 30 degrees Celsius, pH 7.0, and hydraulic retention time 48 hours. The carbon source enrichment rate was obtained by linear regression fitting of volatile suspended solids concentration measured every 12 hours for 72 consecutive hours, and the unit is milligrams per liter per hour. The maximum specific growth rate of photosynthetically active bacteria was obtained by fitting a logistic growth model to five batch culture experiments with different initial substrate concentrations, and the unit is per hour.

[0034] In this embodiment, a gradient boosting regression tree model is trained using carbon source feature vectors as input and photosyntheticity labels as output. Parameters are set as follows: learning rate 0.1, maximum tree depth 5, number of subtrees 100, and minimum number of leaf samples 5. Five-fold cross-validation is used, and a coefficient of determination greater than 0.85 is considered acceptable. After each batch of wastewater treatment is completed, new samples are added to the model for incremental learning and updates. Incremental learning uses an online learning mode, triggering retraining every 50 new samples. New samples have a weight of 1.0, while historical sample weights decrease progressively with a decay coefficient of 0.95. The update frequency is controlled to be once a week to once a month.

[0035] Furthermore, the steps for generating a multi-factor synergistic system include constructing a closed-loop sub-loop of induced factor generation dynamics: A composite photocatalyst was added to a highly active mixed carbon system. The composite photocatalyst was a nano-titanium dioxide supported on iron ions and nitrogen elements and was doped and modified. The amount added was determined by the concentration of dissolved organic carbon in the mixed carbon system through an induction factor demand model. The catalytic reaction was carried out under combined ultraviolet and visible light irradiation conditions, and the intensity sequence of characteristic fluorescence peaks of the inducing factor was monitored in real time by an online three-dimensional fluorescence spectrometer. The characteristic fluorescence peak intensity sequence is input into the inducible factor generation kinetics sub-model, which is constructed based on the photocatalytic oxidation kinetics theory and includes the state variables of hydroxyl radical generation rate and consumption rate. The deviation between the current generation rate and the theoretical maximum generation rate of the induced factor is analyzed in real time using a sub-model of the kinetics of induced factor generation. When the rate deviation exceeds the second preset threshold, the light intensity adjustment is calculated and executed by the proportional-integral-derivative controller based on the rate deviation value, so that the generation rate of the inducing factor is maintained within the preset target range.

[0036] In this embodiment, adding a composite photocatalyst to the highly active mixed carbon system refers to adding a doped and modified catalyst of nano-titanium dioxide supported on iron ions and nitrogen. The catalyst is prepared by the sol-gel method, using tetrabutyl titanate as the titanium source, ferric nitrate as the iron source, and urea as the nitrogen source. The calcination temperature is 450 degrees Celsius for 2 hours. The iron ion doping amount is 0.5% to 2% of the titanium dioxide mass, and the nitrogen doping amount is 1% to 3%. The catalyst has a specific surface area greater than 80 square meters per gram and an average particle size of 20 to 50 nanometers. The catalyst dosage is determined based on the dissolved organic carbon concentration using an inducible factor demand model. This model is constructed using a multiple linear regression algorithm, with the dissolved organic carbon concentration as the input and the catalyst dosage as the output. The model training data comes from 50 batches of wastewater experiments with dissolved organic carbon concentrations ranging from 50 to 500 mg / L. The fitted linear regression equation is that the catalyst dosage equals 0.02 multiplied by the dissolved organic carbon concentration plus 0.3, with a coefficient of determination of 0.92. The dosage range is 0.5 to 3 g / L.

[0037] In this embodiment, the catalytic reaction under combined ultraviolet and visible light irradiation conditions refers to the simultaneous irradiation using an ultraviolet light source with wavelengths of 254 to 365 nm and a visible light source with wavelengths of 400 to 700 nm. The ultraviolet light source is a low-pressure mercury lamp with a power density of 50 to 100 W / m², and the visible light source is an LED array with a power density of 150 to 400 W / m². The ratio of ultraviolet to visible light intensity is 1:3 to 1:5, and the total irradiance is 200 to 500 W / m². The light source is arranged around the perimeter and at the top, with a light intensity non-uniformity of less than 15%. The intensity of the characteristic fluorescence peak of the inducible factor is monitored every minute using an online three-dimensional fluorescence spectrometer (Hitachi F-7100 model), with excitation wavelengths of 200 to 450 nm and emission wavelengths of 250 to 550 nm. The characteristic fluorescence peak corresponds to an excitation wavelength of 280 nm and an emission wavelength of 340 nm.

[0038] In this embodiment, the kinetic sub-model for inducible factor generation is constructed based on photocatalytic oxidation kinetics theory, and the extended Kalman filter algorithm is used to achieve real-time updates of state variables. The model first establishes a set of differential equations based on the photocatalytic oxidation reaction mechanism, including the generation rate of photogenerated electron-hole pairs and the generation and consumption rates of hydroxyl radicals. The kinetic parameters are calibrated using a nonlinear least squares fitting algorithm with 200 sets of historical experimental data. The calibrated set of differential equations is discretized and embedded into the extended Kalman filter framework. The state variables include the generation rate, consumption rate, and concentration of active sites on the catalyst surface; the observed variable is the intensity of the characteristic fluorescence peak. The model inputs are the characteristic fluorescence peak intensity sequence and the current light intensity, catalyst concentration, and dissolved organic carbon concentration. The model outputs the generation rate, consumption rate, and current inducible factor generation rate of hydroxyl radicals.

[0039] In this embodiment, the deviation between the current inducible factor generation rate and the theoretical maximum generation rate is analyzed in real time using a model. The theoretical maximum generation rate is calculated based on a photon yield of 0.05 to 0.1 mol / Einstein, typically ranging from 1.5 x 10⁻⁴ to 3.5 x 10⁻⁴ mol / L per minute. The rate deviation is calculated by subtracting the current generation rate from the theoretical maximum generation rate and then dividing by the theoretical maximum generation rate, converted to a percentage. A second preset threshold is set to 15% to 25%. When the rate deviation exceeds the threshold, the light intensity adjustment is calculated based on the rate deviation value using a proportional-integral-derivative controller. The controller has a proportional coefficient of 0.8, an integral coefficient of 0.2, a derivative coefficient of 0.1, a control cycle of 5 minutes, and an output adjustment range of -30% to +30%. The adjustment is performed by adjusting the light source drive current through a dimmer, with a dimmer control accuracy of ±1%. Monitoring continues after adjustment until the rate deviation returns to within the threshold, maintaining the inducible factor generation rate within 85% to 115% of the theoretical maximum generation rate.

[0040] Furthermore, the harmless transformation steps include constructing a closed loop of dynamic adsorbents for phagocytic factors: During the multi-factor synergistic process, the total radioactivity intensity decay curve of the reaction system is monitored in real time using an online radioactivity detector; The decay curves of viable bacteria and viruses in the reaction system were monitored in real time by combining online flow cytometry with plaque counting. The radioactivity intensity decay curve and the viable bacteria count decay curve were input into the phagocytic factor adsorption kinetics sub-model. The phagocytic factor adsorption kinetics sub-model was constructed based on the Langmuir adsorption isotherm equation and the pseudo-second-order adsorption kinetics equation and included the occupancy rate of the phagocytic factor surface active sites as a state variable. The adsorption efficiency coefficient between the current adsorption rate and the theoretical maximum adsorption rate is analyzed in real time using a phagocytic factor adsorption kinetics sub-model. When the adsorption efficiency coefficient is lower than the third preset threshold, the phagocytic factor replenishment instruction is generated and executed based on the adsorption efficiency coefficient through the fuzzy inference rule base. The input variables of the fuzzy inference rule base are the adsorption efficiency coefficient and the ratio of the current radioactivity intensity to the initial radioactivity intensity.

[0041] In this embodiment, the real-time monitoring of the total radioactivity decay curve of the reaction system during the multi-factor synergistic process using an online radioactivity detector refers to the use of a sodium iodide scintillator detector as the online radioactivity detector. The detector is encapsulated in a stainless steel protective sleeve and directly immersed in the reaction system at a depth of 0.5 meters from the liquid surface. The pulse signal output by the detector is filtered by a single-channel analyzer and then enters the counting circuit, outputting a count rate value per minute, in units of counts per minute. The count rate is compared with a standard radioactive source and converted into a radioactivity intensity value, in units of becquerels per liter, with a detection limit of 0.1 becquerels per liter and a measurement accuracy of ±5%. The monitoring data is automatically uploaded to the control system, and the decay curve of radioactivity intensity over time is plotted in real time to evaluate the adsorption and fixation effect of the phagocytic factors on radionuclides.

[0042] In this embodiment, the real-time monitoring of the viable bacterial and viral count decay curves in the reaction system using a combination of online flow cytometry and plaque counting refers to the combined monitoring using a BD Accuri C6 flow cytometer and double-layer agar plaque counting. The flow cytometer was used to monitor the viable bacterial count. The automatic sample injection rate was 100 μL / min, the sheath fluid was sterile PBS buffer, and SYBR Green I staining combined with propidium iodide double staining was used. Viable bacteria showed green fluorescence, and dead bacteria showed red fluorescence. Excitation was performed using a 488 nm laser. The FL1 channel collected green fluorescence, and the FL3 channel collected red fluorescence. The software automatically calculated the viable bacterial concentration, with a detection limit of 10³ per milliliter. The sampling frequency was once every 30 minutes. Plaque counting was used to monitor the number of viable viruses and bacteria. Samples were taken every 30 minutes, filtered through a 0.22-micron filter to remove bacteria, and the filtrate was mixed with the host bacteria and poured onto double-layer agar plates. After incubation at 37°C for 12 to 18 hours, plaque-forming units were counted, with a detection limit of 10 plaque-forming units per milliliter. Decay curves were plotted on the monitoring data of bacterial viable concentration and viral plaque-forming units over time to evaluate the adsorption and inactivation effect of phagocytic factors on pathogens and viruses.

[0043] In this embodiment, inputting the radioactivity intensity decay curve and viable bacteria count decay curve into the phagocytic factor adsorption kinetics sub-model refers to inputting the radioactivity intensity value sequence, bacterial viable bacteria concentration sequence, and viral plaque formation unit sequence collected every minute into the sub-model. The phagocytic factor adsorption kinetics sub-model is constructed based on the Langmuir adsorption isotherm equation and the pseudo-second-order adsorption kinetic equation, and the model parameters are fitted using the nonlinear least squares method. The model construction process is as follows: First, equilibrium adsorption data at different initial concentrations are obtained through batch adsorption experiments, and the Langmuir adsorption isotherm equation is used for fitting. The equation is: equilibrium adsorption is equal to the maximum adsorption is equal to the Langmuir constant multiplied by the equilibrium concentration divided by 1 plus the Langmuir constant multiplied by the equilibrium concentration. The maximum adsorption is equal to the Langmuir constant and the Langmuir constant are obtained through fitting. Then, the data on the change of adsorption is equal to time are obtained through kinetic experiments, and the pseudo-second-order adsorption kinetic equation is used for fitting. The equation is: adsorption at time t is equal to the pseudo-second-order rate constant multiplied by the square of the maximum adsorption is equal to the t-time multiplied by ... The fitted parameters are used as initial values ​​for the model, and an extended Kalman filter framework is embedded to achieve real-time updates of the state variables. The model input consists of a radioactivity intensity decay sequence and a viable bacteria count decay sequence. The state variables include the occupancy rate of active sites on the phagocytic factor surface, the current adsorption capacity, and the remaining adsorption capacity. The model output is the adsorption efficiency coefficient, which is the ratio of the current adsorption rate to the theoretical maximum adsorption rate. The adsorption efficiency coefficient is defined as the current adsorption rate divided by the theoretical maximum adsorption rate, and its value ranges from 0 to 1. The closer the coefficient is to 1, the higher the adsorption efficiency; the lower the coefficient, the more saturated the adsorption capacity is, requiring the addition of phagocytic factors.

[0044] In this embodiment, the third preset threshold is set to 0.6 to 0.7, with a preferred value of 0.65. When the adsorption efficiency coefficient is lower than the third preset threshold, a phagocytic factor replenishment instruction is generated based on the adsorption efficiency coefficient using a fuzzy inference rule base. The fuzzy inference rule base is constructed using a Mamdani-type fuzzy inference system. The input variables are the adsorption efficiency coefficient and the ratio of the current radioactivity intensity to the initial radioactivity intensity, and the output variable is the phagocytic factor replenishment amount. The universe of discourse for the adsorption efficiency coefficient is set to 0 to 1, divided into three fuzzy sets: low, medium, and high. The membership function uses a triangular function, with the low set center at 0.3, the medium set center at 0.6, and the high set center at 0.9. The universe of discourse for the radioactivity intensity ratio is set to 0 to 1, divided into three fuzzy sets: low, medium, and high. The membership function uses a triangular function, with the low set center at 0.2, the medium set center at 0.5, and the high set center at 0.8. The universe of discourse for the phagocytic factor replenishment amount is set to 0 to 5 grams per liter, divided into five fuzzy sets: zero, small, medium, large, and maximal. The membership function uses a triangular function.

[0045] In this embodiment, the fuzzy rule base contains 9 rules, in the form of "if adsorption efficiency coefficient is A and radioactivity intensity ratio is B then supplementation amount is C". Specifically, the rules are as follows: if the adsorption efficiency coefficient is high and the radioactivity intensity ratio is high, then the supplementation amount is zero; if the adsorption efficiency coefficient is high and the radioactivity intensity ratio is medium, then the supplementation amount is small; if the adsorption efficiency coefficient is high and the radioactivity intensity ratio is low, then the supplementation amount is medium; if the adsorption efficiency coefficient is medium and the radioactivity intensity ratio is high, then the supplementation amount is small; if the adsorption efficiency coefficient is medium and the radioactivity intensity ratio is medium, then the supplementation amount is medium; if the adsorption efficiency coefficient is medium and the radioactivity intensity ratio is low, then the supplementation amount is large; if the adsorption efficiency coefficient is low and the radioactivity intensity ratio is high, then the supplementation amount is medium; if the adsorption efficiency coefficient is low and the radioactivity intensity ratio is medium, then the supplementation amount is large; if the adsorption efficiency coefficient is low and the radioactivity intensity ratio is low, then the supplementation amount is extremely large. Defuzzification uses the centroid method to obtain the precise value of the phagocytic factor supplementation amount, in grams per liter.

[0046] In this embodiment, after the command to replenish the phagocytic factor is executed, a pre-prepared concentrated phagocytic factor solution is added to the reaction system via a metering pump. The concentrated phagocytic factor solution has a concentration of 50 grams per liter and is composed of chitosan-modified material, quaternary ammonium salt functionalized polymer, and zeolite molecular sieve in a mass percentage ratio of 40%, 30%, and 30%, respectively. After replenishment, the adsorption efficiency coefficient is continuously monitored. If the adsorption efficiency coefficient does not recover to above the third preset threshold within 30 minutes, fuzzy inference calculation is performed again and replenishment is repeated until the adsorption efficiency coefficient returns to the normal range. Simultaneously, the triggering conditions and replenishment amount of each replenishment event are recorded and stored in the case library as historical cases for subsequent optimization of the fuzzy rule base.

[0047] Furthermore, the carbon value locking step includes constructing a closed-loop sub-loop that regulates the degree of polymerization of the sugar powder factor: During the curing reaction with the addition of a carbon number locking agent, the particle size distribution sequence of the sugar powder factor polymer was monitored in real time using an online dynamic light scattering instrument; The intensity variation sequence of absorption peaks of characteristic functional groups in the sugar powder factor polymer was monitored in real time using an online Fourier transform infrared spectrometer. The characteristic functional groups include hydroxyl, carbonyl, and ether bonds. The particle size distribution sequence and the intensity variation sequence of the absorption peak of the characteristic functional group are input into the polymerization kinetics sub-model of the sugar powder factor. The polymerization kinetics sub-model of the sugar powder factor is constructed based on the condensation reaction mechanism and includes the state variable of the degree of polymerization distribution. The deviation between the current polymerization progress and the theoretical complete polymerization progress is analyzed in real time using a sugar powder factor polymerization kinetics sub-model. When the progress deviation exceeds the fourth preset threshold, the combined optimization command of the carbon number locking agent replenishment acceleration adjustment amount, stirring speed adjustment amount, and reaction temperature adjustment amount is calculated and executed based on the progress deviation value through the model prediction control algorithm.

[0048] In this embodiment, the real-time monitoring of the particle size distribution sequence of the sugar powder factor polymer during the curing reaction with the addition of a carbon number locking agent refers to the use of a Malvern Zetasizer Pro online dynamic light scattering instrument. The sample cell is equipped with a flow pipeline, and the reaction system sample continuously passes through the sample cell at a flow rate of 5 ml per minute. The laser wavelength is 633 nm, the scattering angle is 173 degrees, and the measurement range is 0.3 nm to 10 μm. Particle size distribution data is automatically collected every 5 minutes, and the output parameters include average particle size, particle size polydispersity index, and particle size distribution curve. The target average particle size range of the sugar powder factor polymer is 1 μm to 10 μm. When the average particle size reaches approximately 5 μm, it indicates that the polymerization reaction has reached the intermediate stage, and a narrowing particle size distribution indicates that the polymerization reaction is becoming more uniform.

[0049] In this embodiment, the real-time monitoring of the absorption peak intensity changes of characteristic functional groups in the sugar powder factor polymer using an online Fourier transform infrared spectrometer refers to employing a Bruker Matrix-F type online Fourier transform infrared spectrometer equipped with an ATR probe directly immersed in the reaction system. The probe is made of diamond, with a wavenumber range of 4000 to 600, a resolution of 4 wavenumbers, 32 scans, and automatic acquisition of infrared spectra every 5 minutes. Characteristic functional groups include a broad absorption peak at 3300 to 3500 for hydroxyl groups, an absorption peak at 1700 to 1750 for carbonyl groups, and an absorption peak at 1050 to 1150 for ether bonds. As the polycondensation reaction proceeds, the intensity of the hydroxyl absorption peak gradually decreases, while the intensity of the carbonyl and ether bond absorption peaks gradually increases. When the intensity of the carbonyl absorption peak stabilizes and no longer changes, it indicates that the polymerization reaction is complete.

[0050] In this embodiment, inputting the particle size distribution sequence and the characteristic functional group absorption peak intensity change sequence into the sugar powder factor polymerization kinetics sub-model refers to inputting the average particle size, particle size polydispersity coefficient collected by a dynamic light scattering instrument, and the characteristic peak intensity value sequences of hydroxyl, carbonyl, and ether bonds collected by a Fourier transform infrared spectrometer into the sub-model. The sugar powder factor polymerization kinetics sub-model is constructed based on the condensation reaction mechanism and uses a model predictive control algorithm to achieve real-time estimation and control of state variables. The model construction process is as follows: First, a set of differential equations for polymerization reaction kinetics is established based on the condensation reaction mechanism, including the equations for the consumption rate of sugar monomers, the equations for the change in the concentration of active end groups, the equations for the evolution of the degree of polymerization distribution, and the equations for the change in system viscosity. Then, the kinetic parameters in the differential equations are calibrated using 200 sets of historical batch experimental data. The parameters include the condensation reaction rate constant, diffusion control coefficient, critical conversion rate at the gel point, etc. The calibration is optimized using a genetic algorithm, with the goal of achieving the highest good fit between the particle size distribution and functional group changes predicted by the model and the experimental data. The calibrated differential equations are discretized and used as the prediction model. State variables include the current degree of polymerization distribution, remaining active end-group concentration, system viscosity, and reaction conversion rate. The model input consists of the online monitored particle size distribution sequence and the characteristic functional group absorption peak intensity change sequence. The state variables are corrected in real time using a Kalman filter algorithm. The model output is the current polymerization progress, expressed as a percentage of reaction conversion rate. The theoretical complete polymerization progress is defined as the state when the reaction conversion rate reaches 98% or higher and the particle size distribution is stable.

[0051] In this embodiment, the progress deviation between the current polymerization progress and the theoretical complete polymerization progress is analyzed in real time through the sugar powder factor polymerization kinetics sub-model. This means comparing the current reaction conversion rate output by the model with the preset theoretical complete polymerization conversion rate of 98%. The progress deviation is equal to the theoretical complete polymerization conversion rate minus the current reaction conversion rate, expressed as a percentage. The progress deviation ranges from 0% to 100%. The larger the deviation value, the further away from the reaction endpoint; a deviation value close to 0% indicates that the reaction is about to be completed.

[0052] In this embodiment, the fourth preset threshold is set to 8% to 12%, with a preferred value of 10%. When the progress deviation exceeds the fourth preset threshold, the model predictive control algorithm calculates a combined optimized instruction for adjusting the carbon number locking agent replenishment rate, stirring speed, and reaction temperature based on the progress deviation value. The prediction time domain of the model predictive control algorithm is set to 60 minutes, the control time domain is set to 30 minutes, and the sampling period is 5 minutes. In each control cycle, the algorithm solves for the optimal control sequence within the next 30 minutes through rolling optimization based on the current state variables and progress deviation. The objective function is to minimize the sum of squared progress deviations, while constraining the variation range of control variables to prevent drastic fluctuations. The control variables include the carbon number locking agent replenishment rate, stirring speed, and reaction temperature, with their respective variation ranges being 0 to 5 g / min, 50 rpm to 300 rpm, and 70°C to 110°C.

[0053] In this embodiment, the carbon-value locking agent acceleration rate is adjusted by regulating the speed of the metering pump, with a flow rate range of 0 to 100 ml / min and a control accuracy of ±2%. The stirring speed is adjusted by regulating the frequency of the stirring motor via a frequency converter. The stirrer is a double-bladed type, with a frequency converter adjustment range of 0 to 50 Hz and a control accuracy of ±0.5 Hz. The reaction temperature is adjusted by regulating the flow rate of the heat transfer oil in the heating jacket. The heating system adopts a cascade structure of PID control and model predictive control, with the inner loop being a fast PID response and the outer loop being the target temperature calculated by model prediction. After the combined optimization command is executed, the polymerization reaction process is monitored by a dynamic light scattering instrument and a Fourier transform infrared spectrometer. If the progress deviation does not decrease to below the fourth preset threshold within 30 minutes, the model predictive control calculation is re-performed and adjusted again until the progress deviation returns to the normal range, ensuring that the sugar powder factor polymer reaches the target degree of polymerization and forms a structurally stable fixed carbon mixture.

[0054] Furthermore, the generation of inducible factors also includes steps for maintaining activity and regulating directional effects: The steady-state concentration of hydroxyl radicals in the inducing factor was monitored in real time using online chemiluminescence. The steady-state concentration of hydroxyl radicals is compared with a preset activity threshold to generate an activity deviation value; When the activity deviation value exceeds the fifth preset threshold, the dissolved oxygen replenishment amount or the co-catalytic aid dosage is calculated by the fuzzy logic controller based on the activity deviation value. The co-catalytic aid is persulfate or hydrogen peroxide. Execute control commands based on the calculated dissolved oxygen replenishment or co-catalytic agent dosage to maintain the oxidative destructive activity of inducing factors on the cell structure of harmful microorganisms. Real-time monitoring of cell morphology changes in harmful microorganisms using an online particle imager yields time-series data on cell damage rate. The time-series data of cell damage rate is compared with the preset target damage rate. When the cell damage rate is lower than the sixth preset threshold, the irradiation intensity of the photocatalytic reaction is automatically increased or the photocatalytic reaction time is extended.

[0055] In this embodiment, real-time monitoring of the steady-state concentration of hydroxyl radicals in the inducing factor using online chemiluminescence refers to employing luminol chemiluminescence. Luminol reagent is continuously injected into the reaction system via a micro-injection pump at a rate of 0.5 mL per minute. The hydroxyl radicals react with luminol to produce chemiluminescence, and the luminescence intensity is detected by a photomultiplier tube (PMT). The PMT operates at 800 volts and is sampled once per minute. The luminescence intensity shows a linear relationship with the hydroxyl radical concentration. A standard curve is established by reacting a series of standard concentrations of hydrogen peroxide with luminol. The standard curve equation is: hydroxyl radical concentration equals luminescence intensity multiplied by 0.002, with a correlation coefficient of 0.995. The detection limit is 1 x 10⁻⁶ mol / L, and the measurement range is from 1 x 10⁻⁶ mol / L to 1 x 10⁻³ mol / L.

[0056] In this embodiment, the preset activity threshold is set to a steady-state concentration of hydroxyl radicals ranging from 1 x 10⁻⁴ mol / L to 5 x 10⁻⁴ mol / L. This threshold is dynamically adjusted based on the dissolved organic carbon (DOC) concentration in the wastewater; for every 100 mg / L increase in DOC concentration, the preset activity threshold is increased by 5%. The real-time monitored steady-state concentration of hydroxyl radicals is compared with the preset activity threshold to calculate the activity deviation value. The activity deviation value is equal to the preset activity threshold minus the measured concentration, divided by the preset activity threshold, multiplied by 100%, and converted to a percentage. The activity deviation value ranges from 0% to 100%.

[0057] In this embodiment, the fifth preset threshold is set to 10% to 20%, with a preferred value of 15%. When the activity deviation value exceeds the fifth preset threshold, the dissolved oxygen replenishment amount or co-catalytic additive dosage is calculated based on the activity deviation value using a fuzzy logic controller. The fuzzy logic controller is constructed using a Mamdani-type fuzzy inference system. The input variables are the activity deviation value and the rate of change of the activity deviation value, and the output variables are the selection and specific value of the dissolved oxygen replenishment amount or the co-catalytic additive dosage. The domain of discourse for the activity deviation value is set to 0% to 100%, divided into three fuzzy sets: small, medium, and large. The membership function uses a triangular function, with the center of the small set at 10%, the center of the medium set at 30%, and the center of the large set at 60%. The domain of discourse for the rate of change of the activity deviation value is set to -10% to +10% per minute, divided into three fuzzy sets: negative, zero, and positive. The membership function uses a triangular function.

[0058] In this embodiment, the fuzzy rule base contains 9 rules, with the rule format being: if the activity deviation value is A and the rate of change of the activity deviation value is B, then the output is C. Specifically, the rules are as follows: if the activity deviation value is small and the rate of change is negative, the output is "small dissolved oxygen replenishment"; if the activity deviation value is small and the rate of change is zero, the output is zero; if the activity deviation value is small and the rate of change is positive, the output is "medium dissolved oxygen replenishment"; if the activity deviation value is medium and the rate of change is negative, the output is "large dissolved oxygen replenishment"; if the activity deviation value is medium and the rate of change is zero, the output is "small persulfate dosage"; if the activity deviation value is medium and the rate of change is positive, the output is "medium persulfate dosage"; if the activity deviation value is large and the rate of change is negative, the output is "large persulfate dosage"; if the activity deviation value is large and the rate of change is zero, the output is "medium hydrogen peroxide dosage"; if the activity deviation value is large and the rate of change is positive, the output is "large hydrogen peroxide dosage". Defuzzification uses the centroid method to obtain the accurate output value.

[0059] In this embodiment, the co-catalytic aid is sodium persulfate, potassium persulfate, or hydrogen peroxide with a mass fraction of 30%. The dosage of persulfate ranges from 0.1 g / L to 0.5 g / L, and the dosage of hydrogen peroxide ranges from 0.1 mL / L to 0.3 mL / L. The dissolved oxygen supplement ranges from 0.5 mg / L to 2 mg / L, achieved by aeration of the reaction system. The aeration flow rate is adjusted by a mass flow controller, with the aeration air flow rate ranging from 0.1 L / min to 1 L / min. Control commands are executed based on the dissolved oxygen supplement or co-catalytic aid dosage calculated by the fuzzy logic controller. The dissolved oxygen supplement is adjusted by the opening of the aeration valve, and the co-catalytic aid dosing rate is adjusted by the metering pump speed. After the control is executed, the steady-state concentration of hydroxyl radicals is continuously monitored to ensure that the steady-state concentration of hydroxyl radicals rises above the preset activity threshold, maintaining the oxidative destructive activity of the inducing factor on the cell structure of harmful microorganisms.

[0060] In this embodiment, real-time monitoring of cell morphology changes of harmful microorganisms using an online particle imager refers to employing a FlowCAM flow cytometer with an automatic sample injection rate of 0.5 ml / min, a flow path thickness of 100 μm, and a high-resolution camera acquiring microbial images at 100 frames per second with a pixel size of 0.5 μm. Real-time analysis of the images using a deep learning image recognition algorithm identifies the proportion of microorganisms with damaged cell membranes to the total number of microorganisms, obtaining time-series data on cell damage rate. The deep learning model uses a convolutional neural network architecture, comprising 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. It is trained using 5000 labeled microbial images, achieving a training set accuracy of 98% and a validation set accuracy of 95%. The sampling frequency is once per minute, outputting a numerical percentage of cell damage rate.

[0061] In this embodiment, the preset target cell breakage rate is set to 85% to 95%, with a preferred value of 90%. The time-series data of cell breakage rate is compared with the preset target breakage rate. When the cell breakage rate is lower than a sixth preset threshold, the irradiation intensity of the photocatalytic reaction is automatically increased or the photocatalytic reaction time is extended. The sixth preset threshold is set to 80% to 90%, which is 5% lower than the preset target breakage rate. The irradiation intensity adjustment range is 10% to 30%, achieved by adjusting the light source drive current, with a dimmer control accuracy of ±1%. The photocatalytic reaction time extension range is 15% to 30%, achieved by extending the reaction residence time. The residence time is controlled by adjusting the feed flow rate, which is adjusted within 70% to 100% of the design flow rate. After the adjustment is executed, the cell breakage rate is continuously monitored until it reaches or exceeds the preset target breakage rate, ensuring that the harmless conversion effect meets the standard.

[0062] In this embodiment, the activity maintenance regulation effect was verified through a control experiment. The control group did not undergo activity maintenance regulation; the initial steady-state concentration of the inducing factor's hydroxyl radicals was 5 x 10⁻⁴ mol / L, decreasing to 0.8 x 10⁻⁴ mol / L after 60 minutes, with a cell rupture rate of 75%. The experimental group employed the activity maintenance regulation of this invention, triggering regulation when the steady-state concentration of hydroxyl radicals fell below 1 x 10⁻⁴ mol / L, with the addition of 0.3 g / L of sodium persulfate. Experimental results: The steady-state concentration of the inducing factor's hydroxyl radicals was maintained between 1.5 x 10⁻⁴ mol / L and 4.5 x 10⁻⁴ mol / L; the effective action time of the inducing factor was extended to 120 minutes; and the cell rupture rate increased to 95%. The improved effect was manifested by a 30% increase in the action time of the inducing factor and a 25% increase in the cell rupture rate.

[0063] Furthermore, the dehydration and drying process includes a biocarbon fuel quality grading and reverse traceability sub-loop: High-purity biocarbon fuel was tested for fixed carbon content, calorific value, ash content, volatile matter, and heavy metal leaching toxicity to obtain multidimensional quality parameters of the biocarbon fuel. The multidimensional quality parameters of biocarbon fuel are compared with the preset target quality parameters to generate a quality deviation vector; The quality deviation vector is input into the quality deviation tracing model based on deep belief network. The hidden nodes of the deep belief network correspond to the key control parameters of each process step. The contribution distribution of each process step to the quality deviation is obtained by backpropagation of the network. Based on the contribution distribution, the key process steps and their key control parameters that lead to quality deviations are identified, and traceability results are generated. The traceability results are fed back to the corresponding process step's sub-closed-loop control system, serving as a reference weight factor for subsequent control of that sub-closed-loop control system.

[0064] In this embodiment, the testing of fixed carbon content, calorific value, ash content, volatile matter, and heavy metal leaching toxicity of high-purity biocarbon fuel refers to sampling and analysis according to the testing methods specified in the relevant national standards. Fixed carbon content testing follows GB / T 28731 standard. One gram of dried biocarbon fuel sample is weighed and placed in a muffle furnace, heated at 900 degrees Celsius for 7 minutes, and the volatile matter content is determined. Then, it is heated at 750 degrees Celsius for 6 hours, and the ash content is determined. Fixed carbon content equals 100 minus volatile matter content minus ash content. Calorific value testing follows GB / T 30727 standard, using an oxygen bomb calorimeter. One gram of sample is weighed and placed in an oxygen bomb, oxygenated to 3 MPa, ignited, and the temperature rise is recorded. The calorific value of the bomb is calculated and converted to the higher heating value. Ash content was determined according to GB / T 30726 standard. One gram of sample was placed in a muffle furnace and heated to 815°C at a heating rate of 10°C per minute, held at that temperature for 2 hours, and the mass of the residue was weighed to calculate the ash content. Volatile matter was determined according to GB / T 30730 standard. One gram of sample was placed in a covered crucible and heated in a muffle furnace at 900°C for 7 minutes. The decrease in mass was weighed to calculate the volatile matter content. Heavy metal leaching toxicity was determined according to HJ 557 standard. One hundred grams of sample was added to one liter of deionized water, the mixture was shaken for 18 hours, filtered, and the concentrations of heavy metals such as lead, cadmium, mercury, arsenic, and chromium in the leachate were determined using inductively coupled plasma mass spectrometry. Multidimensional quality parameters of the biocarbon fuel were obtained, including the percentage of fixed carbon content, calorific value (MJ / kg), percentage of ash content, percentage of volatile matter content, and the leaching concentration of each heavy metal (mg / L).

[0065] In this embodiment, comparing the multidimensional quality parameters of biocarbon fuel with preset target quality parameters involves comparing the measured fixed carbon content, calorific value, ash content, volatile matter content, and heavy metal leaching concentration with the target values ​​specified in the product standard item by item. The preset target quality parameters are set as follows: fixed carbon content greater than 85%, calorific value greater than 25 MJ / kg, ash content less than 10%, volatile matter content less than 5%, lead leaching concentration less than 0.1 mg / L, cadmium leaching concentration less than 0.01 mg / L, mercury leaching concentration less than 0.001 mg / L, arsenic leaching concentration less than 0.05 mg / L, and chromium leaching concentration less than 0.1 mg / L. For each quality parameter, the difference between the measured value and the target value is calculated and divided by the target value to obtain a normalized deviation value. The normalized deviation values ​​of all quality parameters are combined into a multidimensional quality deviation vector with 9 dimensions.

[0066] In this embodiment, inputting the quality deviation vector into the quality deviation tracing model based on a deep belief network means passing the 9-dimensional quality deviation vector as input data to the trained deep belief network. The deep belief network is constructed using a layer-by-layer greedy pre-training and backpropagation fine-tuning approach. The network structure includes an input layer, three hidden layers, and an output layer. The input layer has 9 nodes, corresponding to the deviation values ​​of 9 quality parameters. The first hidden layer has 32 nodes, the second has 16 nodes, and the third has 8 nodes; each hidden layer is pre-trained using a restricted Boltzmann machine. The output layer has 6 nodes, corresponding to the contribution of 6 key process steps, including carbon source enrichment, inducing factor generation, inducing factor activity maintenance, phagocytic factor adsorption, sugar powder factor polymerization, and dehydration / drying. The network training data comes from 500 sets of historical batches of full-process process parameter sequences and corresponding quality deviation vectors, as well as manually labeled key process steps leading to quality deviations. During training, unsupervised pre-training is performed layer by layer, with 100 iterations per layer and a learning rate of 0.1. Then, a Softmax classification layer is added to the output layer for supervised backpropagation fine-tuning, with 200 iterations and a learning rate of 0.01. The loss function is cross-entropy. After training, the hidden nodes of the deep belief network automatically learn the nonlinear mapping relationship between key control parameters of each process step and quality deviations. When the quality deviation vector of the current batch is input, the network calculates the contribution distribution of each process step to the quality deviation through forward propagation. The contribution values ​​range from 0 to 1, and the sum is 1.

[0067] In this embodiment, identifying the key process steps and their key control parameters causing quality deviations based on contribution distribution involves sorting the six contribution values ​​obtained from the output layer from largest to smallest, with the process step having the largest contribution being the key process step causing the quality deviation. When the largest contribution is greater than 0.4 and the second largest contribution is less than 0.3, it is determined to be a single key process step; when the sum of the two largest contributions is greater than 0.7 and the difference between them is less than 0.1, it is determined to be two synergistic key process steps. After determining the key process steps, the sensitivity of the input layer quality deviation vector to the activation values ​​of hidden layer nodes is calculated using a backpropagation algorithm. The hidden layer node with the highest sensitivity corresponds to the key control parameter of that process step. For example, if the carbon source enrichment step is identified as a key process step, the key control parameter affecting this step may be light intensity, trace element supplementation amount, or microbial community compound ratio. A traceability result is generated, including the name of the key process step, the type of key control parameter, the current parameter value, and the recommended adjustment range.

[0068] In this embodiment, feeding back the source tracing results to the corresponding sub-closed-loop control system of the process step means sending the identified key process steps and their key control parameter information to the sub-closed-loop control system corresponding to that step as a reference weight factor for subsequent control. For example, if the source tracing results show that insufficient light intensity in the carbon source enrichment step leads to a low fixed carbon content, this information is fed back to the dynamic control sub-loop of photosynthetic active bacteria, increasing the weight coefficient of the light intensity adjustment in the next control. The weight factor is adjusted by multiplying the weight coefficient of each control measure in the original control strategy library by 1 and adding the source tracing contribution, so that the control measures related to the source tracing results receive higher priority. At the same time, the source tracing results are stored in the historical case library for subsequent incremental updates of the control strategy library and the deep belief network. After every 100 batches of wastewater treatment is completed, the newly added source tracing result data is used to incrementally learn and fine-tune the deep belief network, so that the model continuously adapts to the current water quality characteristics and process operation status.

[0069] Furthermore, the quality deviation tracing model is constructed and updated in the following ways: Collect a large number of historical batches of wastewater treatment process parameters and corresponding high-purity biocarbon fuel multidimensional quality parameters to build a historical case library; For each case in the historical case library, label the actual cause of the quality deviation. The actual cause labels include insufficient carbon source enrichment, insufficient generation of inducing factors, saturation of phagocytic factors, incomplete polymerization of sugar powder factors, and excessively high dehydration and drying temperature. Using the sequence of process parameters throughout the entire process as input and the actual cause labels as output, a gradient boosting decision tree classifier is trained to obtain an initial quality deviation tracing model. After each batch of wastewater treatment is completed, the entire process parameter sequence of that batch and the actual cause labels after manual verification are used as new samples to incrementally learn and update the gradient boosting decision tree classifier.

[0070] In this embodiment, the quality deviation tracing model is constructed using a gradient boosting decision tree classifier. A historical case library is built by collecting a large number of historical batches of wastewater treatment process parameters and corresponding multidimensional quality parameters of high-purity biocarbon fuel. The process parameter sequences include: light intensity, trace element supplementation, microbial community ratio, and hydraulic retention time during the carbon source enrichment stage; catalyst dosage, ultraviolet light intensity, visible light intensity, and reaction temperature during the inducing factor generation stage; steady-state concentration of hydroxyl radicals, dissolved oxygen concentration, and co-catalyst dosage during the inducing factor activity maintenance stage; phagocytic factor dosage, radioactivity attenuation rate, and viable bacteria count attenuation rate during the phagocytic factor adsorption stage; carbon-locking agent dosage, reaction temperature, stirring speed, and pH value during the sugar powder factor polymerization stage; and drying temperature, drying time, and filter press pressure during the dehydration and drying stage. The process parameter sequences for each batch are organized chronologically to form a high-dimensional feature vector. The corresponding multidimensional quality parameters of high-purity biocarbon fuel include fixed carbon content, calorific value, ash content, volatile matter content, and heavy metal leaching concentration. The historical case database has accumulated at least 1,000 batches of data, covering different influent water quality and operating conditions.

[0071] In this embodiment, labeling each case in the historical case database with the actual cause of quality deviation refers to domain experts assigning one or more cause labels to each case based on the correspondence between the entire process parameter sequence and quality parameters, combined with process knowledge and experience. The actual cause labels include five categories: insufficient carbon source enrichment, insufficient generation of inducing factors, saturation of phagocytic factor adsorption, incomplete polymerization of sugar powder factors, and excessively high dehydration drying temperature. Insufficient carbon source enrichment is judged by a carbon source enrichment rate below 0.5 mg / L / h or a maximum specific growth rate of photosynthetic bacteria below 0.1 mg / L / h. Insufficient inducing factor generation is judged by a steady-state concentration of hydroxyl radicals below 1 x 10⁻⁴ mol / L or an inducing factor generation rate below 70% of the theoretical maximum generation rate. Saturation of phagocytic factor adsorption is judged by an adsorption efficiency coefficient below 0.6 and a radioactivity intensity decay rate below 30%. Incomplete polymerization of sugar powder factors is judged by an average particle size of the sugar powder factor polymer less than 1 μm or a carbonyl characteristic peak intensity below 80% of the target value. The criteria for judging excessively high dehydration drying temperature are a drying temperature exceeding 130 degrees Celsius and a decrease in fixed carbon content exceeding 5%. Product batches that meet all quality parameters are marked as qualified batches and are not included in the traceability model training.

[0072] In this embodiment, a gradient boosting decision tree classifier is trained using the entire process parameter sequence as input and the actual cause label as output to obtain the initial quality deviation tracing model. The gradient boosting decision tree classifier is implemented using the XGBoost algorithm, with model parameters set as follows: learning rate 0.1, maximum tree depth 6, number of subtrees 200, minimum number of leaf samples 10, loss function is multi-class log loss, subsampling ratio 0.8, and column sampling ratio 0.8. The input features are standardized numerical vectors of the entire process parameter sequence, with approximately 50 dimensions. The output is the probability values ​​corresponding to 5 categories, and the category with the highest probability is taken as the prediction result. Before training, the historical case library is divided into a training set (80%), a validation set (10%), and a test set (10%). Five-fold cross-validation is used to prevent overfitting. Accuracy, precision, recall, and F1 score are used as model evaluation metrics, and the validation set accuracy is required to be greater than 85% before deployment. Feature importance analysis showed that the most important features for judging insufficient carbon source enrichment were the carbon source enrichment rate and the ratio of bacterial community composition; the most important features for judging insufficient inducing factor generation were the steady-state concentration of hydroxyl radicals and the amount of catalyst added; the most important features for judging adsorption saturation of phagocytic factors were the adsorption efficiency coefficient and the amount of phagocytic factors added; the most important features for judging incomplete polymerization of sugar powder factors were the average particle size and the intensity of carbonyl peaks; and the most important features for judging excessively high dehydration drying temperature were the drying temperature and the fixed carbon content.

[0073] In this embodiment, after each batch of wastewater treatment is completed, the entire process parameter sequence of that batch and the actual cause labels after manual review are used as new samples to incrementally learn and update the gradient boosting decision tree classifier. Manual review is completed by quality inspectors within 24 hours of the quality inspection results. During review, online monitoring data and offline analysis results are combined to confirm the actual causes of quality deviations and label them. Incremental learning adopts an online learning mode, triggering model retraining every 50 new samples. The weight of new samples is set to 1.0, and the weight of historical samples decreases successively with a time decay coefficient of 0.97. That is, the weight of the most recent batch of samples is 1.0, the weight of the previous batch is 0.97, the weight of the batch before that is 0.9409, and so on. During retraining, all historical samples are retained, but the loss function calculation for each sample is multiplied by the corresponding weight coefficient, making the model more focused on the process drift trend reflected by recent data. The model update frequency is controlled at once a month to ensure that the model can continuously adapt to the impact of seasonal changes in water quality, equipment aging, and process improvements. The updated model replaces the original model for subsequent batch quality deviation traceability analysis.

[0074] Furthermore, it also includes a top-level closed loop for end-to-end collaborative optimization based on reinforcement learning: The dynamic regulation sub-loop of photosynthetic active bacteria community, the kinetics sub-loop of inducing factor generation, the regulation sub-loop of inducing factor activity maintenance, the dynamic adsorption sub-loop of phagocytic factor, the polymerization degree regulation sub-loop of sugar powder factor, and the quality grading and reverse traceability sub-loop of biocarbon fuel are jointly constructed into a multi-agent collaborative optimization system. Each sub-loop corresponds to an agent. The action space of each agent is the range of values ​​for its control commands, and the state space of each agent is the feature vector of its real-time monitoring data. Construct a central coordinating agent. The input of the central coordinating agent is the state vector of each sub-closed-loop agent and the initial water quality parameters of the current batch of wastewater. The output of the central coordinating agent is the collaborative weight coefficient of each sub-closed-loop agent. Design a global reward function, which is a weighted sum of the biocarbon fuel yield reward, biocarbon fuel quality reward, unit energy consumption penalty, and processing time penalty. Each sub-closed-loop agent adjusts its own policy network objective function according to the collaborative weight coefficients allocated by the central coordinating agent, and performs collaborative training through a multi-agent deep deterministic policy gradient algorithm to maximize the global reward function value. The multi-agent collaborative optimization system, after training convergence, is deployed in the actual wastewater treatment process to achieve online collaborative optimization and control of the entire process.

[0075] In this embodiment, the construction of a multi-agent collaborative optimization system by integrating the dynamic regulation sub-loop of photosynthetic active microbial community, the kinetics sub-loop of inducible factor generation, the regulation sub-loop of inducible factor activity maintenance, the dynamic adsorption sub-loop of phagocytic factor, the polymerization degree regulation sub-loop of sugar powder factor, and the biocarbon fuel quality grading and reverse traceability sub-loop means that each of the six sub-loops is abstracted into an independent agent, with each agent possessing its own perception, decision-making, and execution capabilities. Agent 1, corresponding to the dynamic regulation sub-loop of photosynthetic active microbial community, is responsible for regulating microbial community activity during the carbon source enrichment stage. Its state space includes the concentration of viable microorganisms, community structure distribution, and carbon source enrichment rate, while its action space includes the adjustment amount of light intensity, the amount of trace element supplementation, and the proportion of microbial community supplementation. Agent 2, corresponding to the kinetics sub-loop of inducible factor generation, is responsible for regulating the generation rate of inducible factors. Its state space includes the intensity of characteristic fluorescence peaks, the generation rate of hydroxyl radicals, and the concentration of dissolved organic carbon, while its action space includes the adjustment amount of light intensity. Agent 3, corresponding to the inducible factor activity maintenance regulation sub-loop, is responsible for maintaining inducible factor activity. Its state space includes the steady-state concentration of hydroxyl radicals and cell damage rate, while its action space includes dissolved oxygen replenishment and co-catalytic agent dosage. Agent 4, corresponding to the phagocytic factor dynamic adsorption sub-loop, is responsible for the adsorption of radioactive substances and pathogens / viruses. Its state space includes radioactivity intensity, viable bacteria count, and adsorption efficiency coefficient, while its action space includes phagocytic factor replenishment. Agent 5, corresponding to the sugar powder factor polymerization degree regulation sub-loop, is responsible for the carbon value locking process. Its state space includes polymer particle size, characteristic functional group peak intensity, and polymerization reaction progress, while its action space includes carbon value locking agent replenishment rate, stirring speed, and reaction temperature. Agent 6, corresponding to the biocarbon fuel quality grading and reverse traceability sub-loop, is responsible for quality detection and traceability feedback. Its state space includes multi-dimensional biocarbon fuel quality parameters and quality deviation vector. Its action space does not directly execute any actions but provides traceability results to other agents as reference weighting factors.

[0076] In this embodiment, constructing a central coordinating agent refers to designing an independent deep neural network as a coordinator, responsible for dynamically allocating the collaborative weight coefficients of each sub-closed-loop agent. The central coordinating agent adopts a multilayer perceptron architecture, with the number of input layer nodes equal to the sum of the dimensions of the state vectors of each sub-closed-loop agent plus the dimension of the initial water quality parameters. The initial water quality parameters include influent COD concentration, dissolved organic carbon concentration, total nitrogen concentration, total phosphorus concentration, types and activities of radionuclides, and viral load, totaling 10 dimensions. The state vector dimensions of each sub-closed-loop agent are as follows: Agent 1: 8 dimensions; Agent 2: 6 dimensions; Agent 3: 4 dimensions; Agent 4: 5 dimensions; Agent 5: 6 dimensions; Agent 6: 9 dimensions, plus the initial water quality parameters (10 dimensions), for a total input dimension of 48 dimensions. The central coordinating agent contains two hidden layers: the first hidden layer has 128 nodes, and the second hidden layer has 64 nodes, both using ReLU activation functions. The output layer has 6 nodes, corresponding to the collaborative weight coefficients of 6 sub-closed-loop agents. The output layer activation function is Softmax, ensuring that the sum of the 6 weight coefficients is 1. The training of the central coordinating agent is carried out simultaneously with the multi-agent reinforcement learning process, with the goal of maximizing the global reward function value.

[0077] In this embodiment, designing a global reward function refers to defining a comprehensive reward index to evaluate the effectiveness of multi-agent collaborative optimization. The global reward function consists of a weighted sum of four terms: biocarbon fuel yield reward, biocarbon fuel quality reward, unit energy consumption penalty, and processing time penalty. The biocarbon fuel yield reward is defined as the ratio of the actual yield to the target yield, where the target yield is set at 80 kg of biocarbon fuel per cubic meter of wastewater. The reward value is 1 when the actual yield exceeds the target yield; otherwise, the reward is the actual yield divided by the target yield. The biocarbon fuel quality reward is defined as the product of fixed carbon content divided by 85% and calorific value divided by 25 MJ / kg. The reward value is 1 when both the fixed carbon content and calorific value are greater than 85% and greater than 25 MJ / kg; otherwise, it is calculated proportionally. The unit energy consumption penalty is defined as the actual energy consumption divided by the reciprocal of the baseline energy consumption, where the baseline energy consumption is set at 200 kWh of electricity per ton of biocarbon fuel. Higher actual energy consumption results in a heavier penalty. The processing time penalty is defined as the standard processing time divided by the actual processing time, with the standard processing time set at 48 hours. Shorter actual processing times result in higher rewards. The weighting coefficients for each item are determined using the analytic hierarchy process (AHP): yield reward weight 0.4, quality reward weight 0.3, energy consumption penalty weight 0.2, and time penalty weight 0.1. The global reward function ranges from 0 to 1; a higher value indicates better collaborative optimization.

[0078] In this embodiment, the objective function of each sub-closed-loop agent's policy network, adjusted according to the collaborative weight coefficients allocated by the central coordinating agent, refers to the calculation of the loss function of the Critic network of each sub-closed-loop agent during the training process of the deep deterministic policy gradient algorithm, which incorporates the weight coefficients allocated by the central coordinating agent. Each sub-closed-loop agent adopts the deep deterministic policy gradient algorithm framework, which includes an Actor network and a Critic network. The Actor network is responsible for outputting actions based on the current state, and the Critic network is responsible for evaluating the value function of the state-action pair. During training, the loss function of each agent's Critic network is multiplied by the corresponding collaborative weight coefficient, so that agents with higher weights receive a larger gradient update magnitude, thereby achieving collaborative optimization. The Actor network structure of agents 1 to 6 is a three-layer fully connected network. The input layer corresponds to their respective state space dimensions, the number of hidden layer nodes are 64 and 32 respectively, and the output layer corresponds to their respective action space dimensions. The activation function is Tanh, which restricts the actions to between -1 and 1, and then maps them to the actual action range. The Critic network structure is a three-layer fully connected network. The input is a concatenation of state and action. The number of hidden layer nodes is 64 and 32 respectively. The output layer outputs the Q value of a single node.

[0079] In this embodiment, collaborative training using a multi-agent deep deterministic policy gradient algorithm refers to employing a framework of centralized training and distributed execution. During training, the Critic network of each agent can acquire global information, while during execution, the Actor network of each agent relies only on local observations. The training process is as follows: First, initialize the parameters of the Actor and Critic networks of the six agents, initialize the network parameters of the central coordinating agent, and initialize the experience replay pool capacity to 10,000 records. In each iteration, randomly sample 256 historical trajectory data from the experience replay pool, including the state, action, reward, and next state of each agent. The central coordinating agent calculates the collaborative weight coefficient based on the current state, and the Critic network of each agent considers the collaborative weight coefficient when calculating the target Q value. The Adam optimizer is used to update the network parameters, with a learning rate of 0.0001 for the Actor network, 0.001 for the Critic network, 0.0005 for the central coordinating agent, and a soft update coefficient of 0.01. The global reward function value is evaluated every 100 rounds during training. Training is considered to have converged when the average reward value of 10 consecutive evaluations is greater than 0.85, and training is stopped.

[0080] In this embodiment, deploying the trained and converged multi-agent collaborative optimization system in an actual wastewater treatment process refers to deploying six trained Actor networks and a central coordinating agent network in the host computer of the industrial control system. At the beginning of each control cycle, each sub-loop collects its own real-time monitoring data to form a state vector, while the water quality analyzer provides the initial water quality parameters for the current batch of wastewater. The central coordinating agent calculates the collaborative weight coefficients based on all state vectors and the initial water quality parameters and sends them to each agent. Each agent's Actor network calculates action commands based on its own state vector, and the action commands are sent to the actuators after amplitude limiting and transformation. The control cycle is set to 15 minutes to ensure that the system can respond quickly to water quality fluctuations. During system operation, new trajectory data is continuously collected and stored in the experience playback pool. Online fine-tuning is triggered every 1000 hours of operation, using the most recently accumulated data to perform small-batch updates to the network parameters, enabling the system to adapt to slow changes during long-term operation. The deployed system achieves online collaborative optimization and control of the entire process, including carbon source enrichment, multi-factor synergistic system generation, harmless conversion, carbon value locking, and dehydration and drying. Under the premise of ensuring biocarbon fuel yield and quality, the unit energy consumption is reduced by 15%, the processing time is shortened by 20%, and the global reward function value is stabilized above 0.88.

[0081] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for converting wastewater into biofuel for power generation, characterized in that, Includes the following steps: Wastewater is introduced into a photosynthetic reactor to enrich carbon sources. Under artificially enhanced photosynthesis, organic carbon, inorganic carbon and trace elements in the wastewater are mixed to form a highly active mixed carbon system. In a highly active mixed carbon system, a multi-factor synergistic system containing inducing factors, phagocytic factors, and sugar powder factors is generated through catalytic reaction. The inducing factors are used to destroy the cell structure of harmful microorganisms, the phagocytic factors are used to fix radioactive materials and pathogens, and the sugar powder factors are used as carbon value locking precursors. A multi-factor synergistic system is used to harmlessly transform wastewater, converting organic pollutants into clean energy carriers of alcohols, while simultaneously immobilizing and capturing radioactive substances and pathogens. After the harmless transformation is completed in the multi-factor synergistic system, the carbon value is locked by a carbon value locking agent to generate a fixed carbon mixture containing phagocytic factor solidified body and sugar powder factor polymer. A mixture of stationary carbons is dehydrated and dried to obtain high-purity biocarbon fuel.

2. The method for converting wastewater into biofuel for power generation according to claim 1, characterized in that, The carbon source enrichment process includes constructing a closed loop for the dynamic regulation of photosynthetic active bacterial communities: The photosynthetic active bacteria are inoculated into the photosynthetic reaction tank. The photosynthetic active bacteria are composed of Chromobacteriaceae, Rhodospirilluae, and Chlorophyticusaceae in a preset ratio. The preset ratio is dynamically determined by a carbon source photosyntheticity prediction model based on the carbon source composition in the wastewater. During the carbon source enrichment process, the number of viable bacteria and the distribution of community structure of photosynthetic active bacteria were monitored in real time by online flow cytometry to obtain real-time monitoring data of bacterial community activity. Real-time monitoring data of microbial community activity is input into the photosynthetic active microbial community metabolic dynamics sub-model, which is constructed based on the principle of microbial quorum sensing and includes state variables of electron transfer flux among microbial communities. The deviation between the current metabolic activity of the photosynthetic microbial community and the expected metabolic pathway is analyzed in real time using a sub-model of the metabolic dynamics of the photosynthetic microbial community. When the deviation exceeds the first preset threshold, the pre-stored control strategy library is matched according to the deviation feature vector to generate and execute dynamic control instructions including light intensity adjustment, trace element supplementation, and microbial community supplementation ratio.

3. The method for converting wastewater into biofuel for power generation according to claim 2, characterized in that, The carbon source photosynthetic capacity prediction model is constructed and dynamically updated in the following manner: We collected a large amount of historical batches of wastewater using three-dimensional fluorescence spectroscopy-high performance liquid chromatography-mass spectrometry full-spectrum data, and extracted fluorescence component scores representing different carbon source components as carbon source feature vectors through parallel factor analysis. The carbon source enrichment rate and the maximum specific growth rate of photosynthetic bacteria under standard photosynthetic induction conditions were collected from each historical batch of wastewater as carbon source photosyntheticity labels. Using carbon source feature vectors as input and carbon source photosyntheticity labels as output, a gradient boosting regression tree model is trained to obtain an initial carbon source photosyntheticity prediction model. After each batch of wastewater is treated, the carbon source feature vector and the measured photosynthetic carbon source data of that batch of wastewater are used as new samples to incrementally learn and update the gradient boosting regression tree model.

4. The method for converting wastewater into biofuel for power generation according to claim 1, characterized in that, The steps for generating a multi-factor synergistic system include constructing a closed-loop sub-loop of the induced factor generation dynamics: A composite photocatalyst was added to a highly active mixed carbon system. The composite photocatalyst was a nano-titanium dioxide supported on iron ions and nitrogen elements and was doped and modified. The amount added was determined by the concentration of dissolved organic carbon in the mixed carbon system through an induction factor demand model. The catalytic reaction was carried out under combined ultraviolet and visible light irradiation conditions, and the intensity sequence of characteristic fluorescence peaks of the inducing factor was monitored in real time by an online three-dimensional fluorescence spectrometer. The characteristic fluorescence peak intensity sequence is input into the inducible factor generation kinetics sub-model, which is constructed based on the photocatalytic oxidation kinetics theory and includes the state variables of hydroxyl radical generation rate and consumption rate. The deviation between the current generation rate and the theoretical maximum generation rate of the induced factor is analyzed in real time using a sub-model of the kinetics of induced factor generation. When the rate deviation exceeds the second preset threshold, the light intensity adjustment is calculated and executed by the proportional-integral-derivative controller based on the rate deviation value, so that the generation rate of the inducing factor is maintained within the preset target range.

5. The method for converting wastewater into biofuel for power generation according to claim 1, characterized in that, The harmless transformation steps include constructing a closed loop of dynamic adsorbents for phagocytic factors: During the multi-factor synergistic process, the total radioactivity intensity decay curve of the reaction system is monitored in real time using an online radioactivity detector; The decay curves of viable bacteria and viruses in the reaction system were monitored in real time by combining online flow cytometry with plaque counting. The radioactivity intensity decay curve and the viable bacteria count decay curve were input into the phagocytic factor adsorption kinetics sub-model. The phagocytic factor adsorption kinetics sub-model was constructed based on the Langmuir adsorption isotherm equation and the pseudo-second-order adsorption kinetics equation and included the occupancy rate of the phagocytic factor surface active sites as a state variable. The adsorption efficiency coefficient between the current adsorption rate and the theoretical maximum adsorption rate is analyzed in real time using a phagocytic factor adsorption kinetics sub-model. When the adsorption efficiency coefficient is lower than the third preset threshold, the phagocytic factor replenishment instruction is generated and executed based on the adsorption efficiency coefficient through the fuzzy inference rule base. The input variables of the fuzzy inference rule base are the adsorption efficiency coefficient and the ratio of the current radioactivity intensity to the initial radioactivity intensity.

6. The method for converting wastewater into biofuel for power generation according to claim 1, characterized in that, The carbon value locking step includes constructing a closed-loop sub-loop that regulates the degree of polymerization of the sugar powder factor: During the curing reaction with the addition of a carbon number locking agent, the particle size distribution sequence of the sugar powder factor polymer was monitored in real time using an online dynamic light scattering instrument; The intensity variation sequence of absorption peaks of characteristic functional groups in the sugar powder factor polymer was monitored in real time using an online Fourier transform infrared spectrometer. The characteristic functional groups include hydroxyl, carbonyl, and ether bonds. The particle size distribution sequence and the intensity variation sequence of the absorption peak of the characteristic functional group are input into the polymerization kinetics sub-model of the sugar powder factor. The polymerization kinetics sub-model of the sugar powder factor is constructed based on the condensation reaction mechanism and includes the state variable of the degree of polymerization distribution. The deviation between the current polymerization progress and the theoretical complete polymerization progress is analyzed in real time using a sugar powder factor polymerization kinetics sub-model. When the progress deviation exceeds the fourth preset threshold, the combined optimization command of the carbon number locking agent replenishment acceleration adjustment amount, stirring speed adjustment amount, and reaction temperature adjustment amount is calculated and executed based on the progress deviation value through the model prediction control algorithm.

7. The method for converting wastewater into biofuel for power generation according to claim 4, characterized in that, After the inducing factor is generated, there are also steps for maintaining its activity and regulating its directional effects: The steady-state concentration of hydroxyl radicals in the inducing factor was monitored in real time using online chemiluminescence. The steady-state concentration of hydroxyl radicals is compared with a preset activity threshold to generate an activity deviation value; When the activity deviation value exceeds the fifth preset threshold, the dissolved oxygen replenishment amount or the co-catalytic aid dosage is calculated by the fuzzy logic controller based on the activity deviation value. The co-catalytic aid is persulfate or hydrogen peroxide. Execute control commands based on the calculated dissolved oxygen replenishment or co-catalytic agent dosage to maintain the oxidative destructive activity of inducing factors on the cell structure of harmful microorganisms. Real-time monitoring of cell morphology changes in harmful microorganisms using an online particle imager yields time-series data on cell damage rate. The time-series data of cell damage rate is compared with the preset target damage rate. When the cell damage rate is lower than the sixth preset threshold, the irradiation intensity of the photocatalytic reaction is automatically increased or the photocatalytic reaction time is extended.

8. The method for converting wastewater into biofuel for power generation according to claim 1, characterized in that, Following the dehydration and drying steps, the process also includes biocarbon fuel quality grading and a reverse traceability sub-loop: High-purity biocarbon fuel was tested for fixed carbon content, calorific value, ash content, volatile matter, and heavy metal leaching toxicity to obtain multidimensional quality parameters of the biocarbon fuel. The multidimensional quality parameters of biocarbon fuel are compared with the preset target quality parameters to generate a quality deviation vector; The quality deviation vector is input into the quality deviation tracing model based on deep belief network. The hidden nodes of the deep belief network correspond to the key control parameters of each process step. The contribution distribution of each process step to the quality deviation is obtained by backpropagation of the network. Based on the contribution distribution, the key process steps and their key control parameters that lead to quality deviations are identified, and traceability results are generated. The traceability results are fed back to the corresponding process step's sub-closed-loop control system, serving as a reference weight factor for subsequent control of that sub-closed-loop control system.

9. The method for converting wastewater into biofuel for power generation according to claim 8, characterized in that, The quality deviation traceability model is constructed and updated in the following ways: Collect a large number of historical batches of wastewater treatment process parameters and corresponding high-purity biocarbon fuel multidimensional quality parameters to build a historical case library; For each case in the historical case library, label the actual cause of the quality deviation. The actual cause labels include insufficient carbon source enrichment, insufficient generation of inducing factors, saturation of phagocytic factors, incomplete polymerization of sugar powder factors, and excessively high dehydration and drying temperature. Using the sequence of process parameters throughout the entire process as input and the actual cause labels as output, a gradient boosting decision tree classifier is trained to obtain an initial quality deviation tracing model. After each batch of wastewater treatment is completed, the entire process parameter sequence of that batch and the actual cause labels after manual verification are used as new samples to incrementally learn and update the gradient boosting decision tree classifier.

10. The method for converting wastewater into biofuel for power generation according to claim 1, characterized in that, It also includes a top-level closed loop for end-to-end collaborative optimization based on reinforcement learning: The dynamic regulation sub-loop of photosynthetic active bacteria community, the kinetics sub-loop of inducing factor generation, the regulation sub-loop of inducing factor activity maintenance, the dynamic adsorption sub-loop of phagocytic factor, the polymerization degree regulation sub-loop of sugar powder factor, and the quality grading and reverse traceability sub-loop of biocarbon fuel are jointly constructed into a multi-agent collaborative optimization system. Each sub-loop corresponds to an agent. The action space of each agent is the range of values ​​for its control commands, and the state space of each agent is the feature vector of its real-time monitoring data. Construct a central coordinating agent. The input of the central coordinating agent is the state vector of each sub-closed-loop agent and the initial water quality parameters of the current batch of wastewater. The output of the central coordinating agent is the collaborative weight coefficient of each sub-closed-loop agent. Design a global reward function, which is a weighted sum of the biocarbon fuel yield reward, biocarbon fuel quality reward, unit energy consumption penalty, and processing time penalty. Each sub-closed-loop agent adjusts its own policy network objective function according to the collaborative weight coefficients allocated by the central coordinating agent, and performs collaborative training through a multi-agent deep deterministic policy gradient algorithm to maximize the global reward function value. The multi-agent collaborative optimization system, after training convergence, is deployed in the actual wastewater treatment process to achieve online collaborative optimization and control of the entire process.

Images