AI-based agricultural multi-waste fermentation nutrient recycling system

Abstract

This invention discloses an AI-based fermentation and nutrient recycling system for diverse agricultural waste, belonging to the field of agricultural waste resource utilization technology. The system includes a pretreatment unit, an AI fermentation unit, and a product processing unit. The pretreatment unit identifies the type and characteristics of the waste through a sensor module and mixes and adjusts the waste according to the proportions calculated by the AI ​​model via a mixing and blending module. The AI ​​fermentation unit monitors the pile status through its sensor system and controls heating, insulation, turning, aeration, and microbial agent supplementation based on the AI ​​model. The product processing unit processes the fermentation products into solid organic fertilizer, organic water-soluble fertilizer, and biochar. This invention solves the problems of reliance on manual experience, crude process parameters, and unstable fermentation efficiency and product quality in traditional agricultural waste treatment. By using an AI model to accurately predict and dynamically control the entire process, it achieves efficient and stable resource utilization and nutrient recycling of waste.

Description

Technical Field

[0001] This invention relates to the field of agricultural waste resource utilization technology, and in particular to an AI-based system for the fermentation and nutrient recycling of diverse agricultural waste. Background Technology

[0002] Agricultural production generates massive amounts of agricultural waste annually, including crop straw, livestock manure, and vegetable scraps. Improper disposal not only wastes resources but can also cause environmental pollution. Therefore, the harmless treatment and resource utilization of agricultural waste are crucial for achieving sustainable agricultural development. Currently, one of the main methods for resource utilization of agricultural waste is aerobic fermentation composting, which aims to convert organic matter into stable humus to produce organic fertilizer, thereby achieving nutrient cycling. However, this technology still faces numerous challenges in large-scale practical processing.

[0003] Agricultural waste comes from a wide range of sources and has a complex composition. Different types of waste, such as straw with high lignocellulose content, vegetable waste with high moisture content and easy decomposition, and livestock and poultry manure rich in nitrogen and phosphorus, exhibit significant differences in key parameters such as physical properties, carbon-nitrogen ratio, and moisture content. Traditional pretreatment methods are often quite extensive, relying on simple mixing or single crushing, making it difficult to achieve precise pretreatment adaptation based on material characteristics. This results in poor uniformity of materials entering the fermentation stage, creating potential problems for subsequent stable fermentation. The fermentation process itself is a complex biochemical process, and its efficiency and product quality are highly dependent on the synergistic effects of many factors, including temperature, oxygen, moisture, and microbial communities. Traditional composting processes rely heavily on fixed schedules and operator experience to control operations such as turning and ventilation, making it difficult to dynamically respond to real-time changes in the internal state of the materials during fermentation. For example, in the low-temperature environment of winter, fermentation is difficult to start, heating is slow, and the cycle is significantly prolonged; while in summer, insufficient ventilation may lead to anaerobic gas production inside the compost pile. This lack of precise control in the process often results in long fermentation cycles, low efficiency, significant nutrient loss, and large fluctuations in the maturity and quality of the final product. The utilization of fermentation products is often limited to the production of primary solid organic fertilizer, failing to achieve refined and high-value nutrient recovery. The leachate produced during fermentation is rich in readily available nutrients and active substances, but it is often treated as wastewater, leading to nutrient loss. Simultaneously, there is a lack of effective utilization pathways for fermentation residues or coarse materials unsuitable for fertilizer production. This singular product route limits the overall efficiency of resource utilization. With increasing attention to the environmental benefits of agricultural production, monitoring and managing energy consumption and greenhouse gas emissions during waste treatment is becoming increasingly necessary. However, existing technologies typically lack the ability to accurately monitor and calculate energy consumption and carbon emissions throughout the entire process, making it difficult to quantify environmental benefits and support further optimization.

[0004] One of the main ways to utilize agricultural waste is through aerobic fermentation composting. However, this technology still faces many technical bottlenecks in large-scale processing. On the one hand, existing AI-controlled technologies only target the single fermentation stage, failing to customize products based on soil-crop needs. While some technologies introduce AI models to regulate the fermentation process, their control logic is limited to the fermentation unit, failing to incorporate the soil nutrient status and crop stage-specific needs of downstream farmland into the AI ​​decision-making model. This results in a disconnect between the fertilizer composition of the fermentation product and the actual needs of the target farmland, hindering precision fertilization and failing to meet the nutrient management requirements of precision agriculture. On the other hand, existing fermentation processes lack quantitative standards for microbial agent addition, leading to low nutrient conversion efficiency. Traditional and some intelligent processes still rely on manual experience or fixed schedules for agent addition, lacking a quantitative control mechanism based on microbial activity feedback. The absence of a dynamic relationship model between enzyme activity and microbial agent supplementation results in large fluctuations in microbial community activity during fermentation, low cellulose degradation rates (generally less than 50%), poor nitrogen conversion efficiency, and easy ammonia volatilization loss, ultimately affecting the nutrient availability of the fermentation product. Furthermore, existing systems do not link carbon emission reductions with carbon asset trading, making it impossible to quantify environmental benefits. Although some existing technologies have begun to focus on monitoring gas emissions during fermentation, a dynamic accounting model for full life-cycle carbon emission reductions has not yet been established, nor has the accounting results been linked to carbon market trading mechanisms (such as CCERs). This prevents the environmental benefits of waste treatment projects from being converted into tradable and quantifiable carbon assets, weakening the economic sustainability and market attractiveness of industrial projects.

[0005] In summary, existing technologies have significant gaps in three aspects: full-process AI closed-loop control, quantitative microbial supplementation, and environmental benefit assetization. This invention addresses these gaps by proposing an AI-based agricultural multi-element waste fermentation nutrient recycling system. This system achieves precise control from pretreatment to product customization through a full-process AI model, and combines a carbon accounting module to quantify and assetize environmental benefits.

[0006] Based on the technical problems mentioned above, there is an urgent need for an agricultural waste treatment technology that can adapt to diverse and complex raw materials, intelligently and precisely control the fermentation process, and realize diversified and high-value utilization of products, in order to improve resource utilization efficiency, stabilize product quality, and enhance environmental friendliness. Summary of the Invention

[0007] This invention overcomes the problems of relying on manual experience, extensive process parameters, and unstable fermentation efficiency and product quality in traditional agricultural waste treatment. By using an AI model to accurately predict and dynamically control the entire process, it achieves efficient and stable resource utilization and nutrient recycling of waste.

[0008] To achieve the above objectives, the present invention adopts the following solution: An AI-based agricultural waste fermentation and nutrient recycling system includes: The pretreatment unit is equipped with an identification sensor module and a mixing and blending module. The identification sensor module identifies the types of agricultural waste and detects its material characteristics. The pretreatment unit performs corresponding crushing or separation treatment according to the type of waste and material characteristics. The mixing and blending module mixes the treated materials with auxiliary materials according to the calculated ratio based on the calculation results of the pre-trained AI model, and adjusts the carbon-nitrogen ratio, moisture content and pH value of the mixture. The AI ​​fermentation unit receives the mixed materials and includes fermentation facilities and its equipped heating and insulation system, sensing system and microbial control system. The sensing system monitors the temperature and oxygen content of the material pile. Based on the monitoring data, the AI ​​fermentation unit controls the operation of the heating and insulation system through an AI model to maintain the fermentation temperature, while controlling the turning and aeration of the pile. The microbial control system inoculates and replenishes compound functional microbial agents into the pile according to the instructions of the AI ​​model. The product processing unit receives the solid materials and fermentation broth obtained after fermentation. The product processing unit includes a solid fertilizer preparation module, a water-soluble fertilizer preparation module, and a biochar preparation module. The solid fertilizer preparation module screens the solid materials, adds trace elements, granulates, and dries them to produce solid organic fertilizer. The water-soluble fertilizer preparation module filters the fermentation broth, uses membrane separation to enrich functional components, and concentrates it to produce organic water-soluble fertilizer. The biochar preparation module pyrolyzes and carbonizes the solid residue to produce biochar and recovers the pyrolysis gas generated during the pyrolysis process.

[0009] As a preferred option, it also includes an AI monitoring and accounting unit and a digital twin unit, which includes a gas monitoring module, a carbon accounting module, and a digital twin platform. The gas monitoring module monitors greenhouse gas emissions and process energy consumption data generated throughout the entire operation of the system. The carbon accounting module calculates the carbon emission reduction of the entire life cycle of the system based on the data obtained by the gas monitoring module through AI algorithms. The digital twin platform constructs a digital twin model synchronized with the physical system, integrates data from the entire system process to simulate and optimize process parameters, and connects farmland soil data and crop growth data to form a closed-loop control.

[0010] Preferably, the gas monitoring module includes gas monitors installed at the inlet and outlet of the fermentation facility, above the pile, and at the tail gas emission outlet of the pyrolysis carbonization furnace, for real-time monitoring of the emission concentrations of methane, carbon dioxide, and nitrous oxide; the gas monitoring module also includes energy consumption monitoring devices installed on waste transport vehicles and each processing unit of the system, for recording transportation fuel consumption and power consumption during the processing. The carbon accounting module is configured to: calculate the system’s carbon emission reduction based on data obtained from the gas monitoring module and energy consumption monitoring device, combined with the CCER methodology for centralized treatment of agricultural waste, using the material balance method and emission factor method. The digital twin platform is built on the Unity3D engine. Its digital twin model is updated synchronously with the physical system at a set frequency. The digital twin platform integrates modules for process parameter optimization, energy consumption monitoring, carbon emission reduction accounting, and product quality traceability. It also accesses data from farmland soil sensors and crop growth monitoring equipment. Based on the accessed data, it uses AI models to simulate and optimize process parameters and adjust the formula and application rate recommendations for fermentation products.

[0011] As a preferred option, the carbon accounting module is also configured to dynamically predict the value of carbon assets and provide trading recommendations, specifically including: The carbon accounting module connects to an external carbon market data platform to obtain real-time information on carbon quota prices and policy changes. Based on historical and real-time data from the gas monitoring module and energy consumption monitoring device, it uses AI algorithms to predict carbon emission reductions within a set future period. The carbon accounting module takes the predicted carbon emission reductions, real-time carbon quota prices, and policy change information as inputs and uses a built-in value assessment model to calculate and generate a future carbon asset value fluctuation curve. When an extreme point that matches the preset trading strategy appears in the calculated carbon asset value fluctuation curve, the carbon accounting module generates a carbon asset trading opportunity suggestion that includes suggested trading time and suggested trading volume, and outputs the suggestion to the digital twin platform for display and early warning.

[0012] As a preferred embodiment, the identification sensing module in the pretreatment unit includes an AI visual recognition device and a multi-parameter sensor array, used to identify the type of agricultural waste and detect its moisture content and carbon-nitrogen ratio. The pretreatment unit performs the following treatments based on the identified waste type: for straw waste, it is graded, crushed, and screened according to its lignification degree; for vegetable waste, it is cleaned and chopped; and for livestock and poultry manure, it is separated into solid and liquid components and dried the solid portion. The mixing and blending module includes a mixer, a quantitative feeder, and a device for adding auxiliary materials. Based on the detection data from the identification sensor module, the mixing and blending module calculates the mixing ratio containing auxiliary materials using an AI model. The quantitative feeder transports the processed waste materials and auxiliary materials to the mixer for mixing according to the mixing ratio. During the mixing process, the mixing and blending module adjusts the carbon-nitrogen ratio, moisture content, and pH value of the mixed materials, monitors material parameters in real time, and dynamically adjusts the feeding amount using an AI model.

[0013] As a preferred option, the fermentation facility in the AI ​​fermentation unit is a fermentation shed with a heat-insulating film, in which the mixed materials are piled up to form a fermentation pile. The heating and insulation system includes solar collectors, phase change heat storage components, and insulation film; the solar collectors are laid on the top of the fermentation shed and connected to the phase change heat storage components through circulation pipes; the insulation film is a multi-layer composite structure. The sensing system includes temperature sensors located at at least two different depths within the stack, and an oxygen sensor located above the stack. The microbial control system includes a storage tank for storing a compound functional microbial agent and a device for applying the agent to the stack. The compound functional microbial agent includes Bacillus subtilis, phosphate-solubilizing bacteria, and nitrogen-fixing bacteria. The AI ​​model is configured to: dynamically control the thermal cycle between the solar thermal collector and the phase change thermal storage components, the opening and closing of the insulation film, the frequency of turning and throwing the pile body and the aeration intensity based on monitoring data from temperature and oxygen sensors; and, based on the assessment of the activity of the microbial community in the pile body, control the microbial regulation system to supplement the pile body with compound functional microbial agents.

[0014] As a preferred method, the specific way to supplement compound functional microbial agents in the AI ​​model-controlled microbial regulation system is as follows: The AI ​​model receives core depth temperature data from the sensor system and cellulase and urease activity data from real-time monitoring of the pile samples. Based on the core depth temperature data, the fermentation process is divided into a warming phase, a high-temperature sustained phase, and a cooling phase. The warming phase is defined as the stage where the core depth temperature rises continuously from the ambient temperature to 55°C; the high-temperature sustained phase is defined as the stage where the core depth temperature remains between 55°C and 60°C; and the cooling phase is defined as the stage where the core depth temperature drops continuously from 55°C to 45°C. During the warming phase, when the cellulase activity falls below a first threshold, a supplementation operation is initiated. The supplementation amount M1 is calculated using the formula M1 = k1 × (A1 - C1), where... k1 is the bacterial replenishment coefficient during the warming period, A1 is the first threshold for cellulase activity, and C1 is the real-time detected cellulase activity value. During the sustained high-temperature period, when the urease activity is lower than the second threshold, the replenishment operation is initiated. The replenishment amount M2 is calculated and determined by the formula M2=k2×(A2-C2), where k2 is the bacterial replenishment coefficient during the high-temperature period, A2 is the second threshold for urease activity, and C2 is the real-time detected urease activity value. The replenished bacterial agents are: during the warming period, mainly the bacterial agent components with cellulose decomposition function in the compound functional bacterial agent; during the sustained high-temperature period, mainly the bacterial agent components with nitrogen conversion function in the compound functional bacterial agent. During the cooling period, when the temperature drops to 45℃ and the activities of both enzymes reach and remain above their corresponding thresholds, the bacterial agent replenishment is stopped.

[0015] Preferably, the solid fertilizer preparation module includes a screening device for screening fermented solid products, an addition device for adding trace elements, a granulator, and a drying device; the solid fertilizer preparation module is configured to: screen the fermented solid products into coarse and fine materials, add trace elements to the fine materials and mix them evenly, granulate the mixed materials, and dry the granulated particles. The water-soluble fertilizer preparation module includes a filtration device, a membrane separation device, a concentration device, and a filling device. The water-soluble fertilizer preparation module is configured to: filter the fermentation liquid product to remove suspended impurities, perform membrane separation on the filtered liquid to enrich functional components, concentrate the enriched liquid, and fill the concentrated liquid. The biochar preparation module includes a pyrolysis carbonization furnace, a cooling device, and a gas treatment device. The biochar preparation module is configured to: transport solid residue to the pyrolysis carbonization furnace for anaerobic pyrolysis to generate biochar and pyrolysis gas; cool and screen the generated biochar; purify the pyrolysis gas; and use the purified gas for power generation or directly provide heat energy to the system.

[0016] Preferably, in the solid fertilizer preparation module, the types and proportions of added micronutrients, as well as in the water-soluble fertilizer preparation module, the target molecular weight cutoff range of the membrane separation device and the final solid content of the concentrate, are determined by an AI model based on soil type data of the target farmland and variety requirements data of the target crop. Soil type data includes soil pH, main nutrient content, and background values ​​of micronutrients; crop variety requirements data includes crop type, growth stage, and historical yield and quality data. The AI ​​model establishes a soil-crop nutrient requirement matching model by analyzing the soil type data and crop variety requirements data. This model first calculates the difference between soil nutrient supply and crop stage requirements, then generates a micronutrient addition formula for the solid fertilizer based on this difference. Simultaneously, it generates a functional component enrichment target for the water-soluble fertilizer based on crop quality improvement requirements. The addition device performs the addition operation according to this micronutrient addition formula, and the membrane separation device in the water-soluble fertilizer preparation module adjusts its operating parameters according to the functional component enrichment target.

[0017] As a preferred approach, the construction and operation of an AI model includes three sequentially executed stages: data fusion, prediction optimization, and instruction generation. The specific construction and operation methods are as follows: During the data fusion phase, the AI ​​model simultaneously receives and integrates multi-source time-series data, including waste type, moisture content and carbon-nitrogen ratio data from the identification sensor module, multi-depth temperature data and oxygen content data of the pile from the AI ​​fermentation unit sensor system, and online detection data of solid product screening efficiency, liquid product functional component concentration and biochar yield from the product processing unit. During the prediction and optimization phase, the core processor of the AI ​​model calls a multi-objective prediction model trained on historical fermentation data. The multi-objective prediction model takes real-time time-series data obtained in the data fusion phase as input and outputs predicted values ​​for the core temperature trend of the pile, the activity trend of key microbial enzymes, and the yield trend of the target product within a set future time period. The AI ​​model compares the predicted values ​​with the preset process target values ​​for each stage and, based on the comparison results, uses an optimization algorithm to solve for a set of controllable process parameter combinations that make the predicted values ​​closest to the process target values. The controllable process parameter combinations include mixing ratio, turning frequency, aeration intensity, microbial agent replenishment amount, and key operating parameter settings for the product processing module. During the instruction generation stage, the AI ​​model converts the controllable process parameter combination obtained in the prediction and optimization stage into a specific sequence of control instructions with time-series markers, and sends them to the actuators in the mixing and blending module, the AI ​​fermentation unit, and the various preparation devices in the product processing module. The AI ​​model is also equipped with a model update mechanism: it associates the final product quality indicators, actual energy consumption data and multi-source time series data of the whole process in each complete fermentation batch to form a new data package, and periodically uses the newly accumulated data package to retrain the multi-objective prediction model to update its model parameters.

[0018] The present invention includes at least the following beneficial effects: (1) Through intelligent identification and allocation of the pretreatment unit, real-time monitoring and dynamic control of the AI ​​fermentation unit, and standardized processing of the product processing unit, the dependence on human experience is significantly reduced, and the stability of system operation and the consistency of processing results are improved; (2) By integrating full-process monitoring, carbon accounting and digital twin technology, and running an AI model of data fusion-prediction optimization-instruction generation, the system can perform forward-looking simulation and optimization, realize adaptive adjustment and closed-loop optimization of process parameters, and greatly improve the overall intelligence level and operating efficiency of the system; (3) By using a solar-biological thermal coupling heating system, the environmental temperature constraint is overcome, ensuring continuous and efficient fermentation; by using a temperature-based The precise microbial agent supplementation strategy based on the degree stage and enzyme activity feedback has achieved targeted enhancement of the microbial community, thereby effectively shortening the fermentation cycle and improving the degradation rate of organic matter and nutrient conversion efficiency; (4) Through the refined pretreatment of different wastes, a foundation for efficient fermentation has been provided; By diverting fermentation products to prepare solid fertilizer, water-soluble fertilizer and biochar, full recovery of nutrients and functional customization of products have been achieved, especially the production of special fertilizers according to soil and crop needs, which greatly enhances the added value and agronomic effectiveness of resource-based products; (5) Through full-chain greenhouse gas monitoring and carbon emission reduction accounting, environmental benefits have been quantified; Furthermore, by predicting the value of carbon assets and providing trading suggestions, the economic sustainability and attractiveness of agricultural projects have been enhanced. Attached Figure Description

[0019] Figure 1 This is a block diagram illustrating the system composition and operating principle of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0021] like Figure 1 As shown, the AI-based agricultural multi-element waste fermentation nutrient recycling system provided by the present invention includes: The pretreatment unit is equipped with an identification sensor module and a mixing and blending module. The identification sensor module identifies the types of agricultural waste and detects its material characteristics. The pretreatment unit performs corresponding crushing or separation treatment according to the type of waste and material characteristics. The mixing and blending module mixes the treated materials with auxiliary materials according to the calculated ratio based on the calculation results of the pre-trained AI model, and adjusts the carbon-nitrogen ratio, moisture content and pH value of the mixture. The AI ​​fermentation unit receives the mixed materials and includes fermentation facilities and its equipped heating and insulation system, sensing system and microbial control system. The sensing system monitors the temperature and oxygen content of the material pile. Based on the monitoring data, the AI ​​fermentation unit controls the operation of the heating and insulation system through an AI model to maintain the fermentation temperature, while controlling the turning and aeration of the pile. The microbial control system inoculates and replenishes compound functional microbial agents into the pile according to the instructions of the AI ​​model. The product processing unit receives the solid materials and fermentation broth obtained after fermentation. The product processing unit includes a solid fertilizer preparation module, a water-soluble fertilizer preparation module, and a biochar preparation module. The solid fertilizer preparation module screens the solid materials, adds trace elements, granulates, and dries them to produce solid organic fertilizer. The water-soluble fertilizer preparation module filters the fermentation broth, uses membrane separation to enrich functional components, and concentrates it to produce organic water-soluble fertilizer. The biochar preparation module pyrolyzes and carbonizes the solid residue to produce biochar and recovers the pyrolysis gas generated during the pyrolysis process.

[0022] The pretreatment unit is responsible for the intelligent identification and pretreatment of input agricultural waste in the system, providing a material basis for subsequent fermentation. The identification sensing module consists of hardware sensors and software algorithms, specifically including a high-resolution camera, a near-infrared spectrometer, a microwave moisture content meter, and a rapid carbon-to-nitrogen ratio (C / N ratio) measuring device. The camera captures images of the waste's appearance and identifies specific types such as straw, livestock manure, and vegetable waste using a pre-trained convolutional neural network model. The near-infrared spectrometer emits light of specific wavelengths and receives the reflected spectrum; through a correlation model between the spectrum and material composition, it non-destructively and accurately detects key characteristics of the material, such as the C / N ratio and lignocellulose content, in real time. The moisture content sensor measures the material's moisture content through changes in dielectric constant or resistance. This detection data is uploaded to the central AI model in real time. Based on the identified waste type and its material characteristics, the pretreatment unit initiates the corresponding physical processing procedures. For example, for straw waste, which is mainly composed of cellulose, hemicellulose, and lignin, the pretreatment unit transports it to a hammer mill or blade mill, where it is crushed into fragments of 1-5 cm in size, depending on the degree of lignification. This size range helps to balance aeration and degradation rate during subsequent fermentation. For vegetable waste, which may contain impurities such as plastic film and soil, the pretreatment unit first removes impurities by passing it through a drum screen, then removes heavy impurities by a hydraulic or pneumatic density sorting device, and finally uses a shredder to homogenize its size to about 2-4 cm. For livestock and poultry manure, the pretreatment unit uses a screw extrusion solid-liquid separator to separate the manure into solid and liquid components at a pressure range of 0.3-0.6 MPa. The solid component is then moderately dried in a belt dryer at a temperature of 60-80 degrees Celsius to adjust its moisture content to a suitable range for subsequent mixing, typically between 50% and 65%. The mixing and blending module includes multiple quantitative feeding hoppers, one or more belt conveyors, and a twin-shaft paddle mixer. After receiving real-time data from the identification sensor module, the AI ​​model uses its internal optimization algorithm, trained on extensive historical data, to calculate an optimal mixing ratio. This ratio not only specifies the weight proportions of the various pre-treated waste materials but also determines the precise amount of auxiliary materials to be added (such as urea and wheat bran for adjusting the carbon-to-nitrogen ratio, or lime and superphosphate for adjusting the pH). The goal of the calculation is to adjust the carbon-to-nitrogen ratio of the final mixture to a range conducive to efficient microbial reproduction, such as 22:1 to 30:1; adjust the moisture content to the range where microbial activity is most active, such as 55% to 65%; and adjust the pH to a neutral or slightly acidic range, such as 6.5 to 7.5. Following instructions from the AI ​​model, the quantitative feeding hoppers deliver various materials and auxiliary materials to the mixer with high precision (e.g., error controlled within ±2%).During the mixing process, online sensors installed at the mixer outlet continuously monitor key parameters of the mixture and feed them back to the AI ​​model. The AI ​​model then dynamically fine-tunes these parameters, forming a closed-loop control system to ensure a high degree of homogeneity in the mixture. Among these modules, the identification sensor module provides perception, the AI ​​model makes decisions, crushing and separating equipment performs pre-processing actions, and the mixing and blending module completes precise synthesis. This transforms complex and diverse raw waste into homogeneous raw materials with stable physicochemical properties, suitable for efficient fermentation.

[0023] The AI ​​fermentation unit, under the precise control of artificial intelligence, transforms the pre-treated mixture into stable humus and nutrients through aerobic fermentation. The AI ​​fermentation unit receives the mixture from the pre-treatment unit and piles it into a geometrically shaped heap within the fermentation facility. The fermentation facility can be a closed or semi-closed fermentation tank with insulation, or a fermentation platform with an openable roof, providing a controllable physical space for microbial fermentation. The unit is equipped with a heating and insulation system to overcome the adverse effects of ambient temperature fluctuations, especially during cold seasons, on the fermentation process. This system can consist of active heating elements (such as a fluid circulation heating loop based on solar collectors or electric heating belts) and passive insulation materials (such as membranes or insulation cotton). By collecting solar energy or utilizing electricity, heat is stored or directly transferred to the fermentation heap, while the insulation layer reduces heat loss, maintaining the interior of the heap within the optimal temperature range for microbial (mainly thermophilic) activity. The sensing system typically includes an array of multiple temperature sensors inserted at different depths within the fermentation pile (e.g., 10 cm, 30 cm, 50 cm below the surface), and oxygen concentration sensors positioned inside or above the pile. These sensors collect data at a fixed frequency (e.g., once every 10 minutes) to reflect the temperature field distribution and oxygen consumption within the pile in real time. The microbial control system is a system comprising a multifunctional microbial agent storage tank, metering pumps, and spraying devices. The stored microbial agent contains various microorganisms that can work synergistically, such as bacteria that decompose cellulose, phosphorus- and potassium-solubilizing bacteria that dissociate phosphorus and potassium, and nitrogen-fixing bacteria that help fix nitrogen. Part of the AI ​​model's operation is as follows: It continuously receives temperature and oxygen data streams from the sensing system. When the temperature data indicates that the core temperature of the pile is below the preset fermentation start-up temperature threshold (e.g., 45 degrees Celsius) or is trending downwards, the AI ​​model sends a command to the heating and insulation system to activate the heating elements or adjust the opening and closing of the insulation film to store heat. Simultaneously, based on oxygen sensor data, if the oxygen content falls below the minimum threshold required for maintaining aerobic fermentation (e.g., volume concentration below 5%), the AI ​​model will determine that the pile is in an anoxic state, triggering two operations sequentially or simultaneously: first, controlling the turning machinery (such as a chain plate turner) to turn the pile, breaking up clumps, loosening the material, and introducing air; second, controlling the aeration blower (such as a Roots blower) to supply air to the air distribution pipes at the bottom of the pile, directly supplementing oxygen. The frequency of turning and the aeration intensity (e.g., air volume per unit time) are both dynamically calculated by the AI ​​model based on real-time data and the fermentation stage. In addition, the AI ​​model also sends instructions to the microbial control system at key time points based on the built-in fermentation stage model (e.g., heating period, high temperature period, cooling period) and microbial activity decay prediction algorithm to perform initial inoculation of microbial agents or supplement microbial agents during fermentation, in order to maintain the activity and diversity of the microbial community.In this unit, the fermentation facility provides the site, the sensing system enables monitoring, the AI ​​model performs analysis, prediction and decision-making, and the heating and insulation system, the turning and aeration actuator, and the microbial control system are responsible for temperature control, oxygen supply control and biological control, respectively. The collaboration of multiple parties ensures that the fermentation process is efficient, stable and controllable.

[0024] The product processing unit is responsible for separating and further processing the fermented material, transforming it into high-value-added products with different forms and uses, achieving full recycling and resource utilization of nutrients. The product processing unit receives the fermented material discharged from the AI ​​fermentation unit, which typically contains a solid component (mature compost) and a liquid component (leachate or fermentation broth). The solid fertilizer preparation module first processes the solid component. The first step is screening, usually using a drum screen or vibrating screen, to classify the solid material according to particle size. For example, coarse, incompletely decomposed components larger than 2 mm are screened out, while the mature, fine material smaller than 2 mm is used as the main material for producing solid organic fertilizer. The screened fine material is conveyed to a mixer, where, according to a preset formula or instructions from the AI ​​model, nutrient salts or chelates of medium-quantity elements (such as calcium, magnesium, and sulfur) and micro-elements (such as iron, zinc, and boron) required for crop growth are added through a precise metering device. The added proportion can be adjusted within a wide range from 1% to 8% according to the needs of the target crop. The uniformly mixed material enters a granulator, where, in a disc granulator or rotary drum granulator, it is processed into granules of uniform size (e.g., 2 mm to 4 mm) by adding an appropriate amount of binder or relying on the material's own adhesive properties. The granules are then fed into a dryer and dried in hot air at 80 to 110 degrees Celsius to reduce their moisture content to a level suitable for storage and transportation (e.g., below 15%). The water-soluble fertilizer preparation module specifically handles the liquid portion. The fermentation broth first undergoes multi-stage filtration, for example, first passing through a grid to remove large particulate impurities, then through a precision filter (such as a bag filter or cartridge filter) to remove fine suspended solids, achieving a filtration accuracy of less than 5 micrometers, resulting in a clear liquid. The clarified liquid then enters a membrane separation system, which typically uses nanofiltration or special ultrafiltration membranes with selectively chosen pore sizes or molecular weight cutoffs (e.g., molecular weight cutoff of 500 to 1000 Daltons). This selectively enriches the functional active ingredients in the liquid, such as small-molecule humic acid, amino acids, and sugars, while allowing water and some inorganic salts to permeate. The concentrated liquid, enriched with functional components, is further concentrated using a vacuum concentration device (such as a rising film evaporator) to increase its solid content to a high level (e.g., above 30%), ultimately producing a high-concentration organic water-soluble fertilizer. The biochar preparation module primarily processes the coarse material screened out during solid fertilizer preparation, as well as solid residues generated in other stages of the system. These materials are fed into a pyrolysis carbonization reactor (such as a continuous rotary kiln), where they are heated to a high temperature (e.g., 400°C to 700°C) in an oxygen-deficient or oxygen-free environment. At this temperature, the organic matter undergoes thermal decomposition; the solid phase product is biochar, and the gaseous phase product is pyrolysis gas (mainly containing combustible gases such as hydrogen, carbon monoxide, and methane). The resulting biochar is cooled, screened, and then packaged into the final product.The pyrolysis gas is then introduced into a purification system to remove tar and dust. It can then be directly fed into the burner to provide heat for the pyrolysis process itself, or used to drive a gas-fired internal combustion engine to generate electricity. The generated electricity can be used by other electrical equipment within the system, thus achieving internal energy circulation. In this unit, the solid fertilizer preparation module achieves solid-phase resource utilization, the water-soluble fertilizer preparation module achieves liquid-phase resource utilization and high-value utilization, and the biochar preparation module achieves residue resource utilization and energy recovery.

[0025] This solution, through the organic integration and synergistic operation of the three units mentioned above, firstly achieves automated and precise pretreatment of diverse agricultural wastes, significantly improving the efficiency and homogenization of raw material processing and creating stable conditions for subsequent fermentation. Secondly, by using artificial intelligence to dynamically regulate multiple parameters such as temperature, oxygen, and microorganisms during the fermentation process, the stability and efficiency of the fermentation process can be greatly improved, the fermentation cycle shortened, and the consistency of the final composted product's quality effectively enhanced. Thirdly, through multi-path, high-value processing and utilization of fermentation products, the resource output rate and economic benefits of the entire system can be significantly improved. Simultaneously, waste is transformed into various forms of agricultural inputs, achieving a closed-loop nutrient cycle. This comprehensively improves the resource utilization level of agricultural waste and has a positive impact on environmental pollution reduction and carbon reduction.

[0026] Another technical solution also includes an AI monitoring and accounting unit and a digital twin unit, which includes a gas monitoring module, a carbon accounting module, and a digital twin platform. The gas monitoring module monitors greenhouse gas emissions and process energy consumption data generated throughout the entire operation of the system. The carbon accounting module calculates the carbon emission reduction of the entire life cycle of the system based on the data obtained by the gas monitoring module and through AI algorithms. The digital twin platform constructs a digital twin model synchronized with the physical system, integrates data from the entire system process to simulate and optimize process parameters, and connects farmland soil data and crop growth data to form a closed-loop control.

[0027] The gas monitoring module tracks greenhouse gas emissions and energy consumption related to waste treatment throughout the entire process. Through dedicated sensors deployed at key nodes, it captures and quantifies changes in the concentrations of greenhouse gases such as methane, carbon dioxide, and nitrous oxide generated during system operation. Simultaneously, energy consumption, including electricity and fuel, is recorded at each stage via energy metering devices. Specifically, the gas monitoring module may include a high-precision infrared gas analyzer, a laser spectrometer, or an electrochemical sensor array. These instruments are strategically installed at critical locations in the system's material and airflow pathways: for example, installation at the inlet and outlet of the fermentation unit monitors changes in the net gas composition emitted during fermentation; installation above the fermentation pile captures gases released from the pile surface; and installation on the exhaust pipe of the pyrolysis carbonization furnace accurately measures gaseous products and potential harmful emissions generated during pyrolysis. The types of gases monitored are not limited to those mentioned; depending on specific processes and accounting requirements, trace gases with environmental impacts, such as ammonia and hydrogen sulfide, may also be included. Energy monitoring devices may include smart meters, fuel flow meters, etc., which are installed on waste transport vehicles to record energy consumption during transportation. They are also installed on major electrical equipment such as pre-processing crushers, mixers, fermentation blowers, and product processing dryers to record the power consumption of each processing stage. All monitoring data is transmitted in real-time to a central data processing system via wired or wireless IoT networks at a high acquisition frequency (e.g., every 5 to 30 minutes). Various sensors distributed in key locations continuously collect raw emission and energy consumption data streams. The data acquisition and transmission network ensures real-time and accurate information upload, providing a high spatiotemporal resolution underlying data foundation for subsequent carbon accounting.

[0028] The carbon accounting module, based on real-time data provided by the gas monitoring module, uses scientific methodologies and algorithmic models to transform complex physicochemical processes into quantifiable and reportable carbon emission reduction indicators. It follows or draws upon internationally recognized carbon emission reduction accounting standards and methodologies, such as those used in the Clean Development Mechanism or nationally certified voluntary emission reductions, to establish a carbon accounting model applicable to the specific process flow of this system. Specifically, this module first cleans, calibrates, and integrates the time-series data uploaded by the gas monitoring module, combining instantaneous concentration data with airflow data to calculate the absolute emissions per unit time at each monitoring point. For energy consumption data, it converts it into carbon dioxide equivalent emissions based on official or authoritative emission factors for different types of energy. The AI ​​algorithm built into the carbon accounting module can perform calculations using a combination of material balance and emission factor methods. The material balance method focuses on calculating carbon fixation and transfer from a material flow perspective, based on the organic carbon content of the input waste and the carbon sequestration amount of the output products (such as stable carbon in biochar). The emission factor method focuses on calculating escape emissions from the perspective of energy consumption and process emissions, based on monitoring data and default factors. The role of the AI ​​algorithm is to intelligently match the optimal accounting path, dynamically adjust parameters, and handle data uncertainties, ultimately calculating the net carbon emission reduction achieved by this system over its entire life cycle compared to traditional waste treatment methods (such as open dumping, landfilling, or incineration). The module's operation is automated: it periodically (e.g., daily or per batch) captures data, executes the accounting model, generates accounting reports, and visualizes key indicators (such as cumulative emission reduction and emission reduction intensity per unit treatment volume).

[0029] A digital twin platform achieves deep understanding, simulation prediction, and forward-looking optimization of the entire process system by constructing a digital twin model that maps and interacts with the physical system in real time. Its construction utilizes technologies such as 3D modeling, real-time data-driven approaches, physicochemical process simulation, and big data analysis to create a digital model in virtual space that is highly consistent with the physical system in terms of geometry, state, behavior, and rules. Specifically, the platform can be developed based on real-time 3D engines such as Unity3D and Unreal Engine, or professional industrial simulation software. The digital twin model not only includes the external 3D models of all equipment, but more importantly, it integrates its control logic, material properties, reaction kinetic parameters (such as microbial growth curves and heat transfer coefficients), and other inherent rules. Through a data interface, it synchronizes real data from the physical system at a high frequency, including material parameters, equipment status (such as valve opening and motor speed), process parameters (temperature, pressure, and gas concentration), and results from gas monitoring and carbon accounting modules. This allows the virtual material state in the virtual model to evolve almost synchronously with the real material state in the physical stack. Building upon this foundation, the digital twin platform integrates data from the entire process, from pretreatment to product processing, forming a unified data lake. Operators or built-in AI algorithms can perform hypothetical analyses within the digital twin model. For example, they can simulate the impact of increasing the initial moisture content by 3 percentage points in winter on the fermentation heating rate, final maturity, and energy consumption; or simulate the impact of adjusting the pyrolysis temperature on biochar yield and carbon sequestration efficiency. By running these simulations, the platform can quickly optimize processes without interfering with actual production, identifying more energy-efficient, more effective, or higher-quality combinations of process parameters, and then distribute the optimized parameter set to the physical system for execution. Furthermore, the platform connects to soil sensor data (such as temperature, humidity, pH, and nitrogen, phosphorus, and potassium content) and crop growth monitoring data (such as leaf index, plant height, and spectral information) from the target farmland via interfaces, thereby streamlining the data flow between the "waste treatment system" and the "farmland production system." Based on this data, the platform can perform reverse optimization: for example, based on the low potassium content in the soil of a certain plot, the functional component enrichment target of the water-soluble fertilizer preparation module can be adjusted to produce high-potassium water-soluble fertilizer in a targeted manner; or the formula ratio of solid organic fertilizer can be adjusted according to the predicted needs of crop growth models.

[0030] This solution upgrades traditional waste treatment facilities into a highly transparent, quantifiable, predictable, and optimizable smart resource factory by integrating panoramic monitoring, intelligent accounting, and high-fidelity digital twin functions into the system. This significantly improves the visualization and reportability of the system's environmental benefits. Simultaneously, through simulation optimization using digital twin technology and closed-loop control based on farmland demand, it effectively enhances the precision of the entire system's operation and the targeting and effectiveness of resource output, thereby significantly improving the overall intelligence level and comprehensive sustainability of the system.

[0031] The gas monitoring module includes gas monitors installed at the inlet and outlet of the fermentation facility, above the pile, and at the tail gas emission outlet of the pyrolysis carbonization furnace, for real-time monitoring of the emission concentrations of methane, carbon dioxide, and nitrous oxide; the gas monitoring module also includes energy consumption monitoring devices installed on waste transport vehicles and each processing unit of the system, for recording transportation fuel consumption and power consumption during the processing. The carbon accounting module is configured to: calculate the system’s carbon emission reduction based on data obtained from the gas monitoring module and energy consumption monitoring device, combined with the CCER methodology for centralized treatment of agricultural waste, using the material balance method and emission factor method. The digital twin platform is built on the Unity3D engine. Its digital twin model is updated synchronously with the physical system at a set frequency. The digital twin platform integrates modules for process parameter optimization, energy consumption monitoring, carbon emission reduction accounting, and product quality traceability. It also accesses data from farmland soil sensors and crop growth monitoring equipment. Based on the accessed data, it uses AI models to simulate and optimize process parameters and adjust the formula and application rate recommendations for fermentation products.

[0032] In practical deployments, gas monitors installed at the inlet and outlet of fermentation facilities typically employ online analyzers to compare the differences in gas composition between the intake and exhaust air, thereby directly calculating the net gas exchange rate during fermentation. For example, they can accurately measure the oxygen consumption rate and carbon dioxide generation rate during fermentation. Monitoring points installed above the pile can utilize open-path laser spectrometers or suspended multi-gas sensor clusters to capture greenhouse gases, particularly methane and nitrous oxide, which diffuse from the pile surface at low concentrations but whose total amount is not negligible. These two gases are easily generated in the local anaerobic environment of the pile. Monitors installed at the exhaust outlet of the pyrolysis carbonization furnace need to withstand higher temperatures and may integrate particulate matter monitoring functions to comprehensively assess the emission characteristics of the pyrolysis process. The core sensors of these gas monitors can employ non-dispersive infrared principles for carbon dioxide, laser spectroscopy or catalytic combustion principles for methane, and high-precision infrared or electrochemical principles for nitrous oxide. The monitoring range and accuracy need to be selected based on the expected emission concentration. For example, the methane monitoring range may be in the range of 0-5000 ppm with an accuracy within ±10 ppm; the nitrous oxide monitoring range is in the range of 0-1000 ppb with an accuracy within ±5 ppb. Regarding energy consumption monitoring devices, those installed on waste transport vehicles typically obtain real-time fuel consumption data through on-board diagnostic systems or external fuel flow meters. Devices installed in each processing unit are mainly smart meters, which are installed in distribution cabinets and measure the electricity consumption of major loads such as crushers, conveyors, fans, water pumps, heaters, and control cabinets on separate circuits. All monitoring devices have remote data transmission capabilities.

[0033] In practical implementation, the carbon accounting module's software program embeds or can call calculation models that meet specific methodological requirements. The material balance method focuses on tracking the flow of carbon elements: for example, the module calculates the total input carbon amount based on the dry matter weight and organic carbon content detection values ​​of various wastes input from the pretreatment unit (which can be provided by the identification sensor module or use typical values); at the same time, based on the output of solid organic fertilizer and biochar and their stable carbon content from the product processing unit, it calculates the amount of carbon fixed and removed from the system in the form of products; the difference between the two, minus the portion lost in the form of carbon dioxide (calculated from gas monitoring data), can be used to estimate the amount of carbon emitted in other forms such as methane. The emission factor rule is applied more directly to energy-related emissions and process emissions: for example, carbon emissions in the transportation stage are obtained by multiplying the diesel consumption (liters) recorded by the energy consumption monitoring device by the carbon dioxide emission factor of diesel (kg CO2 / L); indirect carbon emissions in the processing stage are obtained by multiplying the electricity consumption (kWh) by the emission factor of the regional power grid (kWh CO2 / KWh); for methane and nitrous oxide directly emitted from fermentation and pyrolysis processes, the absolute emissions measured by the gas monitoring module are multiplied by the global warming potential of these two gases relative to carbon dioxide to convert them into carbon dioxide equivalent emissions. The AI ​​algorithm intelligently integrates the calculation results of these two methods, performs cross-validation and uncertainty analysis, automatically handles interpolation or estimation in cases of missing data, and continuously optimizes built-in parameters (such as emission factors for certain specific stages) as data accumulates, making the calculation results increasingly accurate. Specifically, the carbon accounting module, based on data obtained from the gas monitoring module and energy consumption monitoring device, and combined with the National Certified Emission Reduction (CCER) methodology, calculates the system's carbon emission reduction using a combination of material balance and emission factor methods. The specific calculation steps are as follows: 1) Input carbon calculation: Calculate the total input carbon based on the dry matter weight and organic carbon content of various wastes input from the pretreatment unit; 2) Output carbon calculation: Calculate the carbon fixed through products based on the yield of solid organic fertilizer and biochar and their stable carbon content; 3) Emission carbon calculation: Calculate the total greenhouse gas emissions based on the CH4, CO2, and N2O concentrations and airflow measured by the gas monitoring module, and convert them into CO2 equivalents; 4) Carbon emission reduction calculation: The formula is used: Carbon emission reduction = Input carbon - Output carbon - Emission carbon × GWP, where the global warming potential (GWP) of CH4 and N2O are taken as 28 and 265, respectively. The system has built-in typical emission reduction coefficients: fermentation of straw waste: 0.75 t C / t straw, fermentation of livestock and poultry manure: 0.45 t C / t manure, and biochar carbon sequestration: 0.6 t C / t biochar.

[0034] The digital twin platform is built using the Unity3D engine, a real-time 3D development platform widely used in games and industrial visualization. Its advantage lies in its ability to create realistic, smooth, and interactive 3D scenes. The digital twin model is updated synchronously with the physical system at a set frequency (e.g., every 10 minutes, every half hour, or every batch of key nodes). This means that equipment status indicators, material colors (representing temperature or maturity), instrument readings, and even changes in the volume of the virtual pile will be refreshed based on real-time data. Operators can immerse themselves in the entire virtual factory through computers or virtual reality devices. The platform integrates the following functional modules: a process parameter optimization module that uses simulation models for optimization calculations; an energy consumption monitoring module that statistically analyzes and alerts on individual energy consumption items; a carbon emission reduction accounting module that can directly call or display results from an independent carbon accounting module; and a product quality traceability module that uses QR codes or RFID technology to associate each batch of finished fertilizer with its corresponding raw material batch, fermentation process parameters, processing records, and other full lifecycle data, achieving "one item, one code" traceability. The platform not only accesses internal system data but also actively connects to or receives data from external farmland soil sensors (monitoring temperature, humidity, conductivity, pH, and readily available nitrogen, phosphorus, and potassium nutrients) and crop growth monitoring equipment (such as drone multispectral cameras and field video stations) via IoT gateways or data interfaces. The platform's built-in AI model integrates and analyzes this internal and external data. For example, the AI ​​model can run soil-crop nutrient supply and demand simulations. When it predicts that nitrogen in a certain area will be deficient in the next growth stage of the crop, it not only provides fertilization suggestions but also translates this demand into production instructions. This allows it to optimize the membrane separation operation parameters of the water-soluble fertilizer preparation module to enrich a higher proportion of amino acid nitrogen; or adjust the micronutrient addition formula before solid fertilizer granulation to increase the proportion of molybdenum, which promotes nitrogen absorption. These optimized process parameters are then distributed to the corresponding production units through the platform. This solution is based on a mature engine and features a digital twin platform with a clear synchronization mechanism and powerful integration capabilities. It can greatly improve the intuitiveness, predictability and adaptive optimization capabilities of the system operation, and ultimately achieve full-link performance improvement from accurate monitoring and reliable accounting to intelligent decision-making and closed-loop applications.

[0035] The carbon accounting module is also configured to dynamically predict the value of carbon assets and provide trading recommendations, specifically including: The carbon accounting module connects to an external carbon market data platform to obtain real-time information on carbon quota prices and policy changes. Based on historical and real-time data from the gas monitoring module and energy consumption monitoring device, it uses AI algorithms to predict carbon emission reductions within a set future period. The carbon accounting module takes the predicted carbon emission reductions, real-time carbon quota prices, and policy change information as inputs and uses a built-in value assessment model to calculate and generate a future carbon asset value fluctuation curve. When an extreme point that matches the preset trading strategy appears in the calculated carbon asset value fluctuation curve, the carbon accounting module generates a carbon asset trading opportunity suggestion that includes suggested trading time and suggested trading volume, and outputs the suggestion to the digital twin platform for display and early warning.

[0036] The construction of the dynamic carbon asset value prediction function is achieved by accessing external financial markets and policy information flows, combined with the system's continuously generated carbon emission reduction data, and using artificial intelligence algorithms such as time series analysis and regression prediction to make a forward-looking estimate of the potential value of carbon assets over a future period. In specific implementation, the carbon accounting module connects to one or more external carbon market data platforms through encrypted data interfaces. These platforms provide real-time or near-real-time carbon quota trading prices, trading volumes, authoritative announcements (such as adjustments to emission reduction targets, methodological updates, and changes in market access rules), macroeconomic indicators, and other diverse information. The module's internal AI algorithms, such as time series prediction models based on Long Short-Term Memory networks or Transformers, process two sets of data simultaneously: one is the historical and real-time data from the gas monitoring module and energy consumption monitoring devices, which, after accounting, forms a curve of the system's achieved carbon emission reductions; the other is externally accessed carbon price and related policy information. The AI ​​model learns the potential correlation between historical data patterns of carbon emission reduction output and carbon price fluctuations, and considers seasonal factors (such as the seasonality of agricultural waste output) and regulatory shocks to predict the carbon emission reductions that the system may generate within a set period (such as the next quarter, half-year, or compliance year). This prediction period can be flexibly set according to management needs, for example, to the next 12 months to match common carbon market compliance cycles.

[0037] The valuation model comprehensively utilizes financial engineering theories, such as treating predicted carbon emission reductions as a kind of asset spot, and fluctuating carbon prices as a price curve, while introducing concepts such as volatility and risk premium. In its operation, the model uses the predicted future carbon emission reductions obtained from previous forecasts, real-time carbon quota prices, and the sentiment and intensity index of rule change information parsed using natural language processing technology as core inputs. Rule change information, such as a region announcing increased emission reduction efforts, may be quantified as a coefficient that has a positive impact on carbon prices. The model simulates and generates a carbon asset value fluctuation curve covering the future prediction period. This curve not only shows the expected median value but may also demonstrate the possible fluctuation range of value within a certain confidence interval (e.g., a 90% confidence interval) through methods such as Monte Carlo simulation. More importantly, the model internally pre-sets multiple optional trading strategy logics, such as swing trading strategies, hedging strategies based on approaching compliance deadlines, or event-driven strategies based on policy benefits. System administrators can activate one or more of these strategies. When the value fluctuation curve calculated by the model shows an extreme point that conforms to the preset strategy logic (for example, the value curve predicts that it will reach a local peak on the 45th day in the future and then fall back), the model will trigger the suggestion generation mechanism.

[0038] The generation and output of trading recommendations is the final step in transforming internal analysis results into actionable instructions and presenting them to managers. Once the valuation model identifies a suitable trading opportunity (extreme point), the carbon accounting module automatically generates a structured carbon asset trading opportunity recommendation. This recommendation contains at least two core elements: first, the recommended trading time, which typically corresponds to the time window of the predicted value extreme point, providing a specific date or a date range (e.g., "recommend completing the transaction within the next 5-7 days"); second, the recommended trading volume, which is calculated based on factors such as the predicted available carbon emission reductions (i.e., carbon assets) for that period, the company's compliance needs, and risk tolerance. This can be an absolute quantity (e.g., "recommend selling 500 tons of CO2 equivalent") or a percentage of the holdings (e.g., "recommend selling 30% of the current estimated amount"). After generating the recommendation, the carbon accounting module pushes it to the digital twin platform in real time via the internal data bus. The digital twin platform will display this information prominently on its visual interface, such as pop-up notifications on a virtual "carbon asset management center" screen or flashing icons on relevant devices in a 3D scene. Simultaneously, the platform will trigger alerts, which can be achieved through flashing warnings sent via the platform interface, in-system messages to designated operators, or even integrated with external notifications such as SMS, email, or app push notifications, ensuring timely access to critical decision-making information. By endowing the system with carbon asset value prediction and trading decision-making support capabilities, environmental benefit management is elevated to the financial level of asset operation. This significantly enhances project operators' foresight and responsiveness to carbon market opportunities and risks, assisting them in making more scientific and timely trading decisions. Ultimately, this is expected to substantially improve the certainty and optimization of the economic returns generated by the system's environmental rights.

[0039] In another technical solution, the identification sensing module in the pretreatment unit includes an AI visual recognition device and a multi-parameter sensor array, used to identify the type of agricultural waste and detect its moisture content and carbon-nitrogen ratio. The pretreatment unit performs the following treatments based on the identified waste type: for straw waste, it is graded, crushed, and screened according to its lignification degree; for vegetable waste, it is cleaned and chopped; and for livestock and poultry manure, it is separated into solid and liquid components and dried the solid portion. The mixing and blending module includes a mixer, a quantitative feeder, and a device for adding auxiliary materials. Based on the detection data from the identification sensor module, the mixing and blending module calculates the mixing ratio containing auxiliary materials using an AI model. The quantitative feeder transports the processed waste materials and auxiliary materials to the mixer for mixing according to the mixing ratio. During the mixing process, the mixing and blending module adjusts the carbon-nitrogen ratio, moisture content, and pH value of the mixed materials, monitors material parameters in real time, and dynamically adjusts the feeding amount using an AI model.

[0040] The AI ​​visual recognition device consists of an industrial-grade high-resolution color or monochrome camera, supplementary lighting, and an edge computing device or industrial control unit equipped with deep learning algorithms. The camera continuously captures images of waste on the conveyor belt or at the feeding port; the built-in algorithm model, such as a convolutional neural network trained on a large number of labeled images, analyzes the images in real time, identifying different types of waste (such as corn stalks, rice stalks, pig manure, cow manure, tomato vines, cabbage leaves, etc.), and can outline their contours and estimate their volume percentage in the image. A multi-parameter sensor array is a set of physical and chemical sensors installed adjacent to the visual recognition area for rapid detection of the material's non-visual intrinsic properties. This array may include: a near-infrared spectroscopy sensor, which emits near-infrared light of a specific wavelength and analyzes the reflectance spectrum to non-destructively predict key parameters such as moisture content, carbon-to-nitrogen ratio, and lignin content of the material within seconds; a microwave moisture content sensor, which uses the relationship between the material's dielectric constant and moisture content for rapid online measurement; and possibly density sensors, pH probes, etc. These sensors collect data at a frequency of several or dozens of times per second and align the timestamps with the visual recognition results.

[0041] The general processing instructions for the pretreatment unit are specifically divided into three differentiated process paths. For straw-type waste, it is graded and crushed according to its hardness (lignification level), which can be quickly estimated through near-infrared spectroscopy analysis. Based on this data, the control system of the pretreatment unit automatically adjusts the blade gap, speed, or selects different crushing chambers of the crusher. For example, for hard wheat straw with a lignification level higher than a certain threshold (e.g., 55%), the control system instructs the crusher to use a small gap, high speed mode to crush it into shorter fragments, such as 1 to 3 cm in length, to destroy its dense structure and facilitate subsequent microbial contamination. For green corn stalks with a lower lignification level and more flexible texture, a large gap, medium speed mode can be used to crush them into segments of 3 to 5 cm, ensuring the degradation rate while maintaining a certain porosity in the pile. After crushing, a drum screen may be connected to separate excessively fine dust and excessively long uncrushed stalks. For vegetable waste, the core of treatment lies in "impurity removal" and "homogenization." First, the material is guided through a drum-type impurity remover equipped with a magnetic separator and elastic screen, where large foreign objects such as plastic film, ropes, and stones are separated during rotation. Then, the material enters a density separator based on differences in wind or water flow speed, further separating heavy impurities such as mud, sand, and metal fragments with significantly different densities from the vegetable waste. The cleaned vegetable waste is then fed into a chopper, where a high-speed rotating blade breaks it into relatively uniform small pieces, such as 2 to 4 centimeters square, to increase the specific surface area, facilitating subsequent mixing and fermentation. For livestock and poultry manure, the core of treatment lies in "solid-liquid separation" and "moisture content adjustment." The pretreatment unit pumps the waste into a screw extrusion solid-liquid separator. Under mechanical pressure, solid particles are trapped and discharged from the slag outlet, while the liquid (manure slurry) seeps out through the filter screen. The resulting solid fraction typically still has a high moisture content and needs to be gently dried in a paddle dryer or belt dryer at a low temperature (e.g., 60 to 80 degrees Celsius) to reduce its moisture content to a suitable range for direct mixing and blending, such as 50% to 65%, to prevent excessive moisture and clumping. The separated liquid fraction can be temporarily stored in a storage tank or sent to a water treatment unit.

[0042] The core equipment of the mixing and blending module includes: multiple quantitative feeders (such as screw feeders and belt scales) to store various pre-treated waste materials (straw fragments, manure solids, vegetable waste, etc.) and various auxiliary materials (such as urea, rice bran, superphosphate, lime, etc.); a high-efficiency mixer, such as a twin-shaft paddle mixer or a drum mixer; and an online monitoring probe integrated into the mixer outlet or inside (for quickly detecting the moisture content, pH value, etc. of the mixture). When it starts working, the AI ​​model receives detailed characteristic data about the materials in each material bin from the recognition sensor module. Then, based on its internally optimized objectives (such as target C / N ratio, target moisture content, target pH value), combined with preset fermentation process requirements, the model uses linear programming or heuristic algorithms to calculate an optimal mixing ratio, determining how many kilograms of material should be taken from each feeder. Subsequently, the control system instructs each quantitative feeder to deliver the material to the mixer inlet according to this ratio. During the mixer's operation, the online monitoring probe continues to operate. If the real-time monitored parameters of the mixture (such as the moisture content at a certain point) deviate from the theoretical expected value calculated by the AI ​​model, the AI ​​model will immediately initiate a dynamic adjustment program. For example, if the mixture is detected to be too dry, the model will fine-tune by increasing the feed amount of vegetable waste slurry or manure liquid, or activate a micro-spraying device; if the pH value is too low, the amount of alkaline additives will be fine-tuned. This adjustment is continuous and incremental, ensuring that at the end of the mixing cycle, all key indicators of the entire batch of mixture material can stably fall within the preset target range (such as carbon-nitrogen ratio 25:1±1, moisture content 58%±2%, pH value 7.0±0.3). This solution can effectively solve the pretreatment problems caused by the complex composition and diverse properties of agricultural waste. Through the process of identification, classification, and precise allocation, it significantly improves the efficiency and effectiveness of the pretreatment stage, providing crucial raw material support for the stable and efficient operation of subsequent fermentation processes, and improving the treatment efficiency and product quality stability of the entire system from the source.

[0043] In another technical solution, the fermentation facility in the AI ​​fermentation unit is a fermentation shed with a heat-insulating film, in which the mixed materials are piled up to form a fermentation pile. The heating and insulation system includes solar collectors, phase change heat storage components, and insulation film; the solar collectors are laid on the top of the fermentation shed and connected to the phase change heat storage components through circulation pipes; the insulation film is a multi-layer composite structure. The sensing system includes temperature sensors located at at least two different depths within the stack, and an oxygen sensor located above the stack. The microbial control system includes a storage tank for storing a compound functional microbial agent and a device for applying the agent to the stack. The compound functional microbial agent includes Bacillus subtilis, phosphate-solubilizing bacteria, and nitrogen-fixing bacteria. The AI ​​model is configured to: dynamically control the thermal cycle between the solar thermal collector and the phase change thermal storage components, the opening and closing of the insulation film, the frequency of turning and throwing the pile body and the aeration intensity based on monitoring data from temperature and oxygen sensors; and, based on the assessment of the activity of the microbial community in the pile body, control the microbial regulation system to supplement the pile body with compound functional microbial agents.

[0044] The fermentation facility is specifically a fermentation shed with an insulated membrane. This is a semi-open or encloseable structure, with its sidewalls and roof constructed of insulating and translucent or flexible materials. The internal space is used to pile materials to form a fermentation pile. The heating and insulation system is an integrated energy management subsystem that collects, stores, and efficiently utilizes solar energy, supplemented by active insulation methods, to maintain the high-temperature environment required for fermentation. Solar thermal collectors refer to solar air collectors or vacuum tube collectors installed on the roof or south facade of the shed. They absorb solar radiation and convert it into heat energy, heating the working fluid (such as air or antifreeze) flowing through them. Phase change thermal storage components are containers or devices containing specific phase change materials. These materials undergo a solid-liquid phase change at specific temperatures (e.g., around 55-60 degrees Celsius), absorbing or releasing a large amount of latent heat in the process, thus storing heat during periods of ample sunshine and releasing heat at night or on cloudy days. Circulation pipes, pumps, or fans connect the collectors and storage components, forming a closed thermal circulation loop. The multi-layered composite structure of the thermal insulation membrane includes: an outermost layer of weather-resistant, UV-resistant transparent or semi-transparent film (such as fluorocarbon coated film or reinforced polyethylene) that allows sunlight to pass through; a middle layer of high-porosity insulation material (such as non-woven fabric, bubble wrap, or glass wool) that effectively blocks heat conduction; and an inner layer that may be a corrosion-resistant, anti-condensation film. In some specific embodiments, the membrane can also be designed as a rollable or foldable movable structure driven by a motor.

[0045] The sensing system captures key state parameters within the compost heap in real time and in situ. Temperature sensors, positioned at at least two different depths within the heap (e.g., 15 cm below the surface and 50 cm at the core), can map the temperature gradient profile of the heap, which is crucial for determining fermentation uniformity, microbial activity zones, and identifying localized anaerobic overheating. These sensors can be high-temperature resistant platinum resistance thermometers, thermocouples, or digital temperature probes. Oxygen sensors positioned above the heap, typically employing electrochemical or optical principles, monitor the oxygen concentration in gases escaping from the heap surface or in the greenhouse environment, indirectly reflecting the aerobic state within the heap. The microbial control system is a subsystem encompassing the storage, metering, and application of biological agents. Storage tanks for compound functional microbial agents need to be equipped with stirring, insulation (e.g., maintaining 4-10 degrees Celsius), and contamination prevention functions. Devices for applying microbial agents to the heap can be high-pressure spray systems, drip irrigation belts, or solid microbial agent spreaders linked to a turner. In compound functional microbial agents, Bacillus subtilis can produce a variety of enzymes and inhibit harmful bacteria; phosphate-solubilizing bacteria can convert insoluble phosphorus into soluble phosphorus; and nitrogen-fixing bacteria can fix nitrogen from the air. Their combined use can construct an initial microbial community with complementary functions and synergistic effects.

[0046] In this unit, the AI ​​model acts as the central command, coordinating all hardware, responding to sensor data, and driving the system towards the optimal fermentation path. Its configuration logic can be divided into two main control loops. The first loop is the dynamic control of the physical environment. The AI ​​model receives data streams from temperature sensors at different depths in real time. By analyzing this data (such as calculating the heating rate, determining the duration of the high-temperature plateau period, and identifying temperature anomalies), the model dynamically controls the thermal circulation between the solar collector and the phase change storage components. For example, when there is good sunshine during the day but the pile temperature has not reached the target, the model may instruct the circulation pump to start, prioritizing the transfer of heat from the collector to the pile body; when the pile temperature is close to the upper limit and the storage medium is full, the excess heat is stored in the phase change material. At the same time, the model controls the opening and closing of the insulation film based on the temperature difference between the core of the pile and the ambient temperature, closing it at night or in cold weather to maintain heat, and partially opening it during the day when the pile temperature is too high to dissipate heat. Furthermore, by combining oxygen sensor data, the model dynamically determines the frequency of turning (e.g., turning once every 12, 24, or 48 hours) and aeration intensity (e.g., adjusting fan speed to change the ventilation rate per unit time), aiming to balance oxygen supply and heat preservation to maintain an ideal aerobic high-temperature state. The second loop is for the assessment and regulation of the biological community. The AI ​​model not only relies on indirect parameters such as temperature and oxygen, but also integrates rapid detection data from pile samples (e.g., key enzyme activity, total microbial fluorescence value), or utilizes sensor data-driven soft measurement models to assess the overall activity of the microbial community or the abundance of specific functional groups within the pile. When the assessment indicators show that microbial activity is below the threshold required to maintain efficient fermentation, an instruction is sent to the microbial regulation system to control the supplementation of a quantitative amount of compound functional microbial agents to specific areas or the entire pile to reshape or enhance the function of the microbial community. The form of the fermentation facility, the integration method of the heating and insulation system, the layout of the sensor system, and the dual control logic of the AI ​​model can significantly improve the fermentation process's resistance to changes in the external environment and the stability of its internal state, ensuring that the fermentation process can be started quickly and maintained efficiently under different climatic conditions.

[0047] The specific method for supplementing compound functional microbial agents in a microbial regulation system controlled by an AI model is as follows: The AI ​​model receives core depth temperature data from the sensor system and cellulase and urease activity data from real-time monitoring of the pile samples. Based on the core depth temperature data, the fermentation process is divided into a warming phase, a high-temperature sustained phase, and a cooling phase. The warming phase is defined as the stage where the core depth temperature rises continuously from the ambient temperature to 55°C; the high-temperature sustained phase is defined as the stage where the core depth temperature remains between 55°C and 60°C; and the cooling phase is defined as the stage where the core depth temperature drops continuously from 55°C to 45°C. During the warming phase, when the cellulase activity falls below a first threshold, a supplementation operation is initiated. The supplementation amount M1 is calculated using the formula M1 = k1 × (A1 - C1), where... k1 is the bacterial replenishment coefficient during the warming period, A1 is the first threshold for cellulase activity, and C1 is the real-time detected cellulase activity value. During the sustained high-temperature period, when the urease activity is lower than the second threshold, the replenishment operation is initiated. The replenishment amount M2 is calculated and determined by the formula M2=k2×(A2-C2), where k2 is the bacterial replenishment coefficient during the high-temperature period, A2 is the second threshold for urease activity, and C2 is the real-time detected urease activity value. The replenished bacterial agents are: during the warming period, mainly the bacterial agent components with cellulose decomposition function in the compound functional bacterial agent; during the sustained high-temperature period, mainly the bacterial agent components with nitrogen conversion function in the compound functional bacterial agent. During the cooling period, when the temperature drops to 45℃ and the activities of both enzymes reach and remain above their corresponding thresholds, the bacterial agent replenishment is stopped.

[0048] The continuous fermentation process is specifically divided into three periods with clearly defined temperature boundaries: the warming phase, the sustained high-temperature phase, and the cooling phase. This division is based on the classic theory of thermophilic aerobic composting and provides precise numerical boundaries. The data source for this division is the core depth temperature data from the sensing system, which refers to the temperature sensor readings located at the geometric center of the compost or the most representative depth (e.g., 50-60 cm). The warming phase is defined as the stage from the start of fermentation, where the core temperature of the compost rises continuously from the ambient temperature until it first reaches a critical threshold (e.g., 55°C). This threshold (e.g., 55°C) marks the shift from mesophilic microbial dominance to thermophilic microbial dominance, signifying the entry of fermentation into a high-speed phase. The sustained high-temperature phase is defined as the stage where the core temperature of the compost remains within a certain high-temperature range (e.g., between 55°C and 60°C). This range is the ideal interval where most harmful pathogens and weed seeds are inactivated, while the activity of the thermophilic microbial community is at its highest. The cooling period is defined as the stage where the core temperature of the fermentation pile begins to decrease from the lower limit of the high-temperature period (e.g., 55 degrees Celsius) until it reaches another lower threshold (e.g., 45 degrees Celsius). A temperature of around 45 degrees Celsius may mark the end of primary fermentation and the beginning of secondary aging or the product stage. In practical applications, these temperature thresholds (e.g., 55°C, 60°C, 45°C) are typical reference values. Depending on the specific materials and inoculant characteristics, similar values ​​such as 50°C, 58°C, and 40°C may also be used as dividing points.

[0049] Two key biochemical feedback indicators are introduced here: cellulase activity and urease activity, with thresholds (first threshold A1 and second threshold A2) set to trigger supplementation operations. Cellulase activity reflects the ability of microorganisms to degrade lignocellulosic recalcitrant organic matter and is crucial for initiating material decomposition during the heating phase. Urease activity reflects the ability of microorganisms to convert nitrogen-containing organic matter (such as urea and protein) and is closely related to nitrogen preservation and conversion, especially important for nitrogen regulation during the high-temperature phase. Real-time detection of these enzyme activities typically requires automated sampling from the stockpile using a portable enzyme activity analyzer or a rapid detection kit based on colorimetric reactions. During the heating phase, when the real-time detected cellulase activity value C1 is lower than the preset first threshold A1 = 15 U / g (range 10-20 U / g), the AI ​​model determines that its own microbial cellulose decomposition capacity is insufficient and thus initiates a supplementation operation. The replenishment amount M1 is calculated using a linear compensation formula, with units of kg / t. It represents the mass of microbial agent required per ton of material. The principle is that the replenishment amount is proportional to the difference between the target activity (threshold A1) and the actual activity (C1). The microbial agent replenishment coefficient k1 during the heating period in the formula is an empirical parameter that can be determined experimentally. It takes into account factors such as the effectiveness of the microbial agent and the size of the stockpile, and its value ranges from 0.05 to 0.2 kg / (t·U / g). It represents the mass of microbial agent required per ton of material for every 1 U / g difference in enzyme activity, reflecting the intensity of demand for cellulose-decomposing bacteria during the heating period. A1 is the first threshold for cellulase activity, and C1 is the real-time detected value of cellulase activity. During the sustained high-temperature period, the logic is similar, but the monitoring target becomes urease activity C2. When it falls below the second threshold A2 = 12 U / g (range 8-15 U / g), supplementation is initiated. The supplementation amount M2 is calculated using a formula incorporating the high-temperature period coefficient k2, with k2 ranging from 0.08-0.25 kg / (t·U / g). Similar to k1, k2 reflects the intensity of nitrogen-converting bacteria demand during the high-temperature period. A2 is the second threshold for urease activity, and C2 is the real-time detected urease activity value. The supplemented bacterial agent composition has different focuses: during the warming period, the supplemented bacterial agent mainly consists of components with cellulose decomposition function in compound functional bacterial agents, such as those containing a higher proportion of lignocellulose-degrading bacteria (e.g., specific actinomycetes or fungi); during the sustained high-temperature period, the supplemented bacterial agent mainly consists of components with nitrogen conversion function, such as those containing a higher proportion of nitrogen-solubilizing, nitrogen-fixing, or nitrifying bacteria. The above threshold values ​​are based on the correlation between microbial enzyme activity and nutrient conversion efficiency in plant nutrition: when cellulase activity is below 10 U / g, the straw cellulose degradation rate decreases by more than 30%; when urease activity is below 8 U / g, nitrogen conversion efficiency is significantly reduced, which easily leads to ammonia volatilization loss.

[0050] At the end of fermentation, when the temperature drops to a specific point (e.g., 45 degrees Celsius) and the activities of both key enzymes reach and remain above their corresponding thresholds, the inoculum replenishment is stopped. A temperature of 45 degrees Celsius indicates that the intense biodegradation phase of fermentation has essentially ended, and the process has entered a stable maturation phase. At this point, if the activities of cellulase and urease have reached and remained stable, it indicates that the decomposition and transformation of organic matter have been fully completed, and no further inoculum is needed to drive the reaction. The phrase "reach and remain above their corresponding thresholds" requires the model to determine that the activity values ​​within a time window are consistently above the threshold to avoid misjudgments due to fluctuations in a single test. The command to stop replenishment is issued by the AI ​​model, and the microbial control system then enters standby mode. This scheme achieves extremely precise, on-demand, timed, quantitative, and component-specific control of the microbial community during fermentation by precisely and quantitatively coupling the biological operation of inoculum replenishment with the fermentation temperature stage and key enzyme activity indicators. This control method goes beyond empirical control based on time or a single temperature. It can significantly improve the efficiency and targeting of microbial agents, effectively address the challenges brought about by fluctuations in different batches of materials, and ensure that the fermentation process always maintains efficient and targeted metabolic capacity at the microbiological level, thereby fundamentally improving the fermentation rate, stability, and consistency of product quality.

[0051] In another technical solution, the solid fertilizer preparation module includes a screening device for screening fermented solid products, an addition device for adding trace elements, a granulator, and a drying device; the solid fertilizer preparation module is configured to: screen the fermented solid products into coarse and fine materials, add trace elements to the fine materials and mix them evenly, granulate the mixed materials, and dry the granulated particles. The water-soluble fertilizer preparation module includes a filtration device, a membrane separation device, a concentration device, and a filling device. The water-soluble fertilizer preparation module is configured to: filter the fermentation liquid product to remove suspended impurities, perform membrane separation on the filtered liquid to enrich functional components, concentrate the enriched liquid, and fill the concentrated liquid. The biochar preparation module includes a pyrolysis carbonization furnace, a cooling device, and a gas treatment device. The biochar preparation module is configured to: transport solid residue to the pyrolysis carbonization furnace for anaerobic pyrolysis to generate biochar and pyrolysis gas; cool and screen the generated biochar; purify the pyrolysis gas; and use the purified gas for power generation or directly provide heat energy to the system.

[0052] The task of the solid fertilizer preparation module is to process the fermented solid materials into commercial solid organic fertilizers with uniform particles, balanced nutrients, and easy storage and application. The process begins with screening, typically using a drum screen or vibrating screen. The screen aperture diameter can be selected according to the required particle size of the finished product, for example, set to 2 mm, 3 mm, or 4 mm. Through screening, the material is separated into coarse material (particle size larger than the screen aperture) and fine material (particle size less than or equal to the screen aperture). The coarse material is usually returned to the pretreatment unit or sent to the biochar preparation module for further processing, while the fine material is used as the main raw material for solid fertilizer production. Next, trace elements are added. The addition device can be a micro-scale batching scale linked to the main material conveyor belt, or a premixing bin with a stirring function. There are many types of micronutrients that need to be added, including medium elements such as calcium, magnesium, and sulfur, as well as trace elements such as iron, zinc, boron, and molybdenum. The proportion of added micronutrients is usually in the range of one percent to a few percent of the total material. The types and proportions of micronutrients added are determined by the AI ​​model based on the soil type data of the target farmland and the variety requirements data of the target crop. Depending on the crop type, the AI ​​model automatically matches micronutrient supplementation strategies. For example: 1) For fruit and vegetable crops (such as tomatoes, cucumbers, and apples), the focus is on supplementing calcium (Ca), boron (B), and zinc (Zn) at proportions of 2%-5%, 0.5%-1%, and 1%-2%, respectively, to promote fruit enlargement, prevent fruit cracking, and increase sugar content; 2) For grain crops (such as rice, corn, and wheat), the focus is on supplementing magnesium (Mg), silicon (Si), and molybdenum (Mo) at proportions of 1%-3%, 2%-4%, and 0.1%-0.3%, respectively, to enhance lodging resistance, improve photosynthetic efficiency, and promote grain filling; 3) For legume crops (such as soybeans and peanuts), the focus is on supplementing molybdenum (Mo) and cobalt (Co) at proportions of 0.1%-0.2% and 0.02%-0.05%, respectively, to promote nitrogen fixation in root nodules. These matching strategies are based on crop nutrient absorption patterns in plant nutrition and soil nutrient abundance / deficiency indicators, ensuring that fertilizer formulations are targeted and agronomically effective. Based on the formulation determined by the AI ​​model, these elements are precisely added in the form of sulfates, oxides, chelates, etc., and thoroughly mixed with the fine materials in the main mixer. The mixed material then enters a granulator, commonly a disc granulator, rotary drum granulator, or extrusion granulator. In a disc granulator, the material gradually forms dense spherical particles within a rotating, tilting disc by spraying an appropriate amount of water or binder. The particle size can be controlled by adjusting the disc angle, rotation speed, and residence time, for example, to obtain particles of 2 to 4 millimeters. The wet granules after granulation have a high moisture content and need to be dried. The drying device can be a rotary drum dryer, fluidized bed dryer, or belt dryer. The hot air temperature needs to be carefully controlled, for example, within the range of 80 to 110 degrees Celsius, to avoid high temperatures damaging organic matter and effective microorganisms.The goal of drying is to reduce the moisture content of the granules to a low and stable level, such as below 15%, to ensure that the product does not become moldy during storage.

[0053] The water-soluble fertilizer preparation module is specifically designed to process fermentation broth rich in soluble nutrients produced during fermentation, aiming to produce highly concentrated, highly active liquid organic water-soluble fertilizer. The first step is filtration. The filtration device can employ a multi-stage series configuration, for example, first passing the broth through a vibrating screen or drum screen to remove large suspended fibers, then proceeding to a plate and frame filter press or bag filter for fine filtration. The filtration precision can reach 5 microns or even lower to remove most suspended impurities and colloids, resulting in a clarified fermentation broth. Next is membrane separation to enrich functional components. The core of the membrane separation device is a selective separation membrane, such as a nanofiltration membrane with a molecular weight cutoff in the range of 500 Daltons to 2000 Daltons. Under pressure, small molecules (such as water and inorganic salts) permeate through the membrane as the "permeate," while larger target functional molecules (such as humic acid, proteins, and polysaccharides) are retained as the concentrate. Through this process, the functional components are enriched several times over. The operating parameters of the membrane system, such as pressure, flow rate, and temperature, need to be optimized and controlled, for example, the operating pressure should be between 1.0 MPa and 2.5 MPa. Although the concentration of functional components in the enriched liquid is increased, its total volume is still relatively large, requiring further dehydration in a concentration unit. The concentration unit can employ a low-energy multi-effect evaporator or a mechanical vapor recompression evaporator, concentrating the liquid to a high solids content state, for example, a final solids content of 30% to 50%, under relatively low evaporation temperatures (e.g., 60 to 70 degrees Celsius) and vacuum conditions. Finally, the concentrate is dispensed quantitatively into plastic drums or flexible packaging of different sizes, completing the commercial packaging.

[0054] The preparation and application of biochar are based on its soil-improving effect: biochar has a porous structure and high specific surface area, which can enhance the soil's water and fertilizer retention capacity, adsorb and fix available nutrients in the soil, and reduce leaching losses. Simultaneously, the stable carbon components in biochar can remain in the soil for a long time, achieving carbon sequestration, which aligns with the concept of "synergistic improvement of nutrient cycling and soil quality" in plant nutrition. The biochar preparation module uses thermochemical conversion to produce biochar with stable carbon fixation and soil-improving functions from the solid residues generated in the system (such as screened coarse materials and non-granular fibers), while also achieving energy recovery. The core equipment is a pyrolysis carbonization furnace, which can be a continuously fed externally heated rotary kiln, a fluidized bed furnace, or an intermittent fixed-bed reactor. In an oxygen-deficient or anaerobic environment, the solid residues are heated to relatively high temperatures (e.g., 400°C to 700°C), causing the organic matter to undergo thermal decomposition rather than complete combustion. During this process, the solid material is converted into biochar rich in stable carbon, while releasing volatiles to form pyrolysis gas. The pyrolysis process requires precise control of the heating rate and final temperature to balance the yield, properties, and energy recovery of biochar. The resulting biochar is produced at extremely high temperatures and needs to be cooled to near ambient temperature using cooling devices such as water-cooled spirals, jacketed cooling cylinders, or inert gas countercurrent cooling towers. The cooled biochar then undergoes sieving to obtain uniform particle size. The pyrolysis gas is a mixture of combustible gases, primarily composed of hydrogen, carbon monoxide, methane, carbon dioxide, and some tar and moisture. Gas treatment devices are used to purify these gases, for example, by removing dust through cyclone separators and removing tar and moisture through condensers and electrostatic precipitators. The purified clean pyrolysis gas has a high calorific value and can be utilized in two ways: one is to directly introduce it into the burner to provide the heat energy required to maintain the reaction temperature of the pyrolysis carbonization furnace itself, thereby significantly reducing the external energy demand; the other is to introduce it into a small gas internal combustion engine or gas turbine to generate electricity, which can supply other electrical equipment in the system, such as fans, water pumps, lighting and control cabinets, to achieve energy self-sufficiency or partial self-sufficiency.

[0055] The three product processing paths set in this scheme can ensure that the system can fully and wastelessly recover the fermentation products, output organic matter and nutrients in the most suitable forms for agricultural applications (solid fertilizer, liquid fertilizer, soil conditioner), and significantly reduce the overall energy consumption and operating costs of the system through internal energy recycling, thereby greatly improving the economy, environmental friendliness and industrial applicability of the entire resource recycling system.

[0056] In the solid fertilizer preparation module, the types and proportions of added micronutrients, as well as in the water-soluble fertilizer preparation module, the target molecular weight cutoff range and final solid content of the concentrate in the membrane separation device are determined by an AI model based on soil type data of the target farmland and variety requirements data of the target crop. Soil type data includes soil pH, main nutrient content, and background values ​​of micronutrients; crop variety requirements data includes crop type, growth stage, and historical yield and quality data. The AI ​​model establishes a soil-crop nutrient requirement matching model by analyzing the soil type data and crop variety requirements data. This model first calculates the difference between soil nutrient supply and crop stage requirements, then generates a micronutrient addition formula for the solid fertilizer based on this difference. Simultaneously, it generates a functional component enrichment target for the water-soluble fertilizer based on crop quality improvement requirements. The addition device performs the addition operation according to this micronutrient addition formula, and the membrane separation device in the water-soluble fertilizer preparation module adjusts its operating parameters according to the functional component enrichment target.

[0057] The key to customized decision-making lies in the types and proportions of micronutrients in solid fertilizers, as well as the target and concentration endpoint of water-soluble fertilizer film separation. These are determined by AI models based on two types of external data: soil type data of the target farmland and variety requirements data of the target crop. Soil type data is a set of parameters describing the basic properties and nutrient status of the soil, typically obtained through soil testing reports or online field sensors. It includes, but is not limited to, soil pH (acidic, neutral, or alkaline), the content of major nutrients (such as available nitrogen, available phosphorus, and available potassium), and the background values ​​of micronutrients (such as exchangeable calcium and magnesium, and available iron, zinc, and boron). Crop variety requirements data reflects the nutrient requirements of crops throughout their growth cycle. This includes crop type (such as corn, tomato, and apple), growth stage (such as seedling stage, vegetative growth stage, flowering stage, and fruiting stage), and yield and quality data based on historical records or targets (such as target sugar content, vitamin content, and fruit size). The AI ​​model analyzes this data and executes the core algorithm of a soil-crop nutrient demand matching model. Based on the principle of nutrient supply and demand balance in plant nutrition, this model calculates the difference between available soil nutrients and the crop's stage-specific requirements. A customized nutrient formulation is initiated when the difference is ±10 mg / kg. Soil supply is based on detected baseline values, while crop requirements are derived from a built-in database of crop nutrient absorption models. This difference (positive surplus or negative deficiency) precisely indicates the types and quantities of nutrients that need to be supplemented or adjusted through fertilization.

[0058] After calculating the difference between soil supply and crop demand, the soil-crop nutrient requirement matching model first generates specific instructions for solid fertilizer production. For micronutrients, the model focuses on analyzing the difference between soil background values ​​and crop requirements. For example, if a farmland is diagnosed with low soil pH (acidic), and the crop (fruit tree) has a high calcium requirement during fruit enlargement, but the soil's available calcium background value is insufficient, the model will determine that calcium supplementation is needed. It will not only decide to add calcium but also calculate a specific addition ratio based on the severity of calcium deficiency (the size of the difference) and the crop's sensitivity, such as recommending adding 3% to 5% calcium conditioner (e.g., oyster shell powder or calcium nitrate) to the solid fertilizer formulation. Simultaneously, the model will also consider the synergistic or antagonistic effects of other elements, ultimately generating a "solid fertilizer micronutrient addition formula" containing multiple micronutrients and their precise proportions. This formula is a structured data list specifying each substance to be added and its target percentage in the final mixed dry matter. The additive device (i.e., micro-addition system) in the solid fertilizer preparation module will strictly follow this formula list to perform the addition operation.

[0059] In addition to solid fertilizers, the demand-matching model also guides the production of water-soluble fertilizers based on crop quality improvement needs. Crop quality improvement is a higher-level objective, potentially stemming from contract farming requirements for sweetness, color, and storability. The model associates these quality goals with specific functional components. For example, to improve the sugar content and flavor of tomatoes, the model might link it to supplementing amino acids and certain trace elements; to improve the aroma of tea, it might link it to enriching specific organic acids. Based on this, the model generates a "functional component enrichment target for water-soluble fertilizers." This target is not a simple formulation but a series of requirements for the types and concentrations of functional components in the final water-soluble fertilizer product. This target is assigned to the water-soluble fertilizer preparation module, where the membrane separation device is the key execution unit for achieving this target. The operating parameters of the membrane separation, especially its "target molecular weight cutoff range," directly determine which molecules of different sizes and properties will be enriched. For example, to enrich small-molecule humic acid and amino acids with a molecular weight of around 1000 Daltons, the model will instruct the membrane system to use a nanofiltration membrane with the corresponding cutoff range. In addition, the final solids content target of the concentrate also needs to be set to ensure that the finished product has a sufficient concentration of active substances. Therefore, the AI ​​model (or the module controller that receives the model's instructions) will dynamically adjust the operating parameters of the membrane separation device, such as system pressure, circulation flow rate, selection of membrane modules or cleaning cycle, according to the functional component enrichment target, and set the target value for the concentration endpoint.

[0060] Customized production of fermentation products is based on soil fertility grading standards and crop nutrient absorption patterns, following the law of minimum in plant nutrition. The AI ​​model analyzes the difference between the available soil nutrient content and the crop's stage-specific requirements to generate precise formulas, enabling a "supplement what's lacking" nutrient management strategy. For example, when the available zinc content in the soil is below a critical value (<0.5 mg / kg) and the crop is corn, the AI ​​model automatically adds zinc to the solid fertilizer at a ratio of 1.5%-2.5% to eliminate limiting factors and increase yield. This solution achieves a fundamental shift from "applying what is produced" to "producing what is needed" by intelligently linking the fertilizer preparation process with front-end farmland soil and crop demand data. Since the ultimate goal of fertilizer preparation is to benefit crops, if the fertilizer composition cannot be adjusted to meet the needs of the target crop, the fertilization effect cannot be optimized, and fertilizer failure may even occur. This solution can significantly improve the targeting and agronomic effectiveness of the produced fertilizer products, ensuring a high degree of matching between nutrient input and crop needs in terms of time, space, and type. This can better support high-yield and high-quality crop production while reducing blind fertilization and nutrient waste, greatly enhancing the core value and competitiveness of the entire system in precision agriculture and green agriculture.

[0061] In another technical solution, the construction and operation of the AI ​​model includes three sequentially executed stages: data fusion, prediction optimization, and instruction generation. The specific construction and operation method is as follows: During the data fusion phase, the AI ​​model simultaneously receives and integrates multi-source time-series data, including waste type, moisture content and carbon-nitrogen ratio data from the identification sensor module, multi-depth temperature data and oxygen content data of the pile from the AI ​​fermentation unit sensor system, and online detection data of solid product screening efficiency, liquid product functional component concentration and biochar yield from the product processing unit. During the prediction and optimization phase, the core processor of the AI ​​model calls a multi-objective prediction model trained on historical fermentation data. The multi-objective prediction model takes real-time time-series data obtained in the data fusion phase as input and outputs predicted values ​​for the core temperature trend of the pile, the activity trend of key microbial enzymes, and the yield trend of the target product within a set future time period. The AI ​​model compares the predicted values ​​with the preset process target values ​​for each stage and, based on the comparison results, uses an optimization algorithm to solve for a set of controllable process parameter combinations that make the predicted values ​​closest to the process target values. The controllable process parameter combinations include mixing ratio, turning frequency, aeration intensity, microbial agent replenishment amount, and key operating parameter settings for the product processing module. During the instruction generation stage, the AI ​​model converts the controllable process parameter combination obtained in the prediction and optimization stage into a specific sequence of control instructions with time-series markers, and sends them to the actuators in the mixing and blending module, the AI ​​fermentation unit, and the various preparation devices in the product processing module. The AI ​​model is also equipped with a model update mechanism: it associates the final product quality indicators, actual energy consumption data and multi-source time series data of the whole process in each complete fermentation batch to form a new data package, and periodically uses the newly accumulated data package to retrain the multi-objective prediction model to update its model parameters.

[0062] During the data fusion phase, the AI ​​model perceives the physical world and obtains entry points for decision-making materials, simultaneously receiving and integrating multi-source heterogeneous data with time-series characteristics from different stages of the system. This data must be precisely aligned and fused in both time and logic. First, data from the identification sensor module in the pretreatment unit includes waste type identification results (e.g., code codes), real-time moisture content (e.g., percentage values), and carbon-nitrogen ratio (e.g., ratio value). This data is typically uploaded several times per minute or per batch of material processed. Second, the data stream from the AI ​​fermentation unit's sensing system is more intensive, including readings from multiple temperature sensors deployed at different depths within the pile (e.g., surface, middle, and core layers), and oxygen content data above the pile. This data may be collected every 5 to 15 minutes, collectively depicting the dynamic microenvironment within the pile. Finally, online detection data from the product processing unit includes efficiency data during solid product sieving (e.g., fine material yield), functional component concentrations obtained from sampling and analysis of liquid products (e.g., humic acid content), and real-time estimation data of biochar yield. These data may be generated hourly or per batch. To achieve effective fusion, all data is automatically timestamped with high precision upon arrival and transmitted via a unified data bus or message middleware. The data fusion module within the AI ​​model cleans, aligns, and interpolates this data according to a unified timeline (such as absolute system time or relative fermentation batch time), eliminating outliers and integrating data from different frequencies into a unified time-series snapshot or sequence for use in the next stage. As an example, during the data fusion stage, the AI ​​model synchronously receives and integrates multi-source time-series data. All data is aligned and fused with timestamps of 1 minute precision to ensure the synchronization of multi-source data on the timeline. For different units, the multi-source time-series data includes: 1) Pre-processing unit: waste type, moisture content, carbon-nitrogen ratio (sampling frequency: 1 time / minute); 2) AI fermentation unit: multi-depth temperature of the pile, oxygen content (sampling frequency: 1 time / 5 minutes); 3) Product processing unit: screening efficiency, functional component concentration, biochar yield (sampling frequency: 1 time / hour).

[0063] The prediction and optimization phase determines how the AI ​​model, based on current and historical state data, predicts future trends and solves for the optimal operational strategy. Essentially, it employs a multi-objective prediction model pre-trained on a large amount of historical fermentation data. This model can be constructed using advanced time-series prediction algorithms such as Long Short-Term Memory (LSTM) networks and gated recurrent units (GRUs), capable of capturing the complex nonlinear dynamics of the fermentation process. During operation, the model uses the integrated real-time time-series data from the previous phase (data fusion phase) as its primary input. This input data can include temperature sequences, oxygen content trends, and initial material properties over the past few hours. Depending on the size of the training dataset required by the model, it typically collects more than 100 batches of complete fermentation data. Based on these inputs, the multi-objective prediction model is configured to simultaneously output predicted values ​​for multiple key objectives within a set future time period. This prediction time period can be set as needed, for example, predicting trends for the next 12, 24, or 48 hours. Predicted objectives typically include: the changing trend of the core temperature of the fermentation pile, the activity trend of key enzymes reflecting microbial activity (such as cellulase and protease), and the expected yield trend of target products (such as high-quality solid fertilizer and high-concentration water-soluble fertilizer). After obtaining these predicted values, the AI ​​model compares them with pre-set process target values ​​expected to be achieved at each fermentation stage. These process target values ​​can be a range, such as a target temperature range of 55-60 degrees Celsius during the high-temperature stage, and cellulase activity not falling below a certain threshold. Next, the AI ​​model calls its built-in optimization algorithm (such as a genetic algorithm, particle swarm optimization, or gradient-based optimizer) to search for solutions within the allowed operating space, with the optimization objective of "making the predicted values ​​as close as possible to or reaching the process target values." This solution process outputs a set of optimized and controllable process parameters, covering the entire chain from pretreatment to product processing. Specifically, this includes: the mixing ratio of various materials and auxiliary materials in pretreatment; the operating frequency of the turning and turning machinery during fermentation (e.g., once every 8 hours) and the intensity setting of the aeration blower (e.g., air volume in cubic meters per hour); the timing and amount of microbial agent replenishment; and the setting values ​​of key operating parameters in the product processing module, such as drying temperature and membrane separation pressure.

[0064] During the instruction generation phase, the AI ​​model needs to transform the abstract combination of controllable process parameters obtained in the prediction and optimization phase into a specific, executable sequence of control instructions with precise timing markers. For example, it not only includes the specific operations of the instructions but also specifies the start time (e.g., starting at 14:00 today) and duration. These instructions are packaged and distributed to the corresponding physical execution units: mixing ratio instructions are distributed to the mixing and blending module of the pretreatment unit; instructions regarding turning, aeration, and inoculum replenishment are distributed to the corresponding actuators (turning machine, blower, inoculum pump) in the AI ​​fermentation unit; and instructions regarding drying temperature and membrane pressure are distributed to the various preparation devices in the product processing module. These instructions are transmitted through an industrial control network to ensure that the actuators perform the correct actions at the correct time. Beyond the execution of individual decisions, a crucial model update mechanism is employed, endowing the AI ​​model with the ability to learn and continuously improve. Specifically, the system meticulously correlates the actual results data obtained from each complete fermentation batch with multi-source time-series data collected throughout the fermentation process, encapsulating this into a structured new data package. The actual results data includes product quality indicators (such as organic matter content, nutrient content, and impurity rate) and the actual energy consumption data for the entire process. As the system continues to operate, these new data packages accumulate. The system periodically (e.g., every 10 batches or once a month) uses these newly accumulated data packages to retrain the multi-objective prediction model. During retraining, new data is used to adjust the model's internal parameters (such as the weights of the neural network), enabling the model to better learn and adapt to new material properties, environmental conditions, or equipment states encountered in actual operation, thereby continuously improving its prediction accuracy and optimization performance. This complete, closed-loop workflow of AI model perception-thinking-action-learning is deeply and organically integrated into the complex physical and biological processing system, endowing the system with powerful real-time state perception and prediction capabilities, allowing it to proactively adjust operations rather than passively responding. By solving multi-objective optimization problems, the optimal balance can be found among mutually constraining process objectives (such as speed, quality, and energy consumption). Finally, combined with the model's continuous self-updating mechanism, the entire system can be ensured to exhibit high adaptability, robustness, and continuously evolving optimization capabilities during long-term operation, thereby significantly improving the intelligence level, stability, and overall efficiency of the entire resource recovery process.

[0065] To further demonstrate some of the beneficial effects of the present invention in practical application, the following comparative test process and results are provided: Experimental grouping and conditions: Control group 1: Traditional artificially regulated fermentation process (manual turning, no AI regulation, no customized product processing); Control group 2: Existing AI single-stage regulated fermentation process (AI regulation only for fermentation temperature and turning, aseptic dosage supplementation, no product customization); Experimental group: The whole-process AI-regulated fermentation process of this invention (including AI pretreatment ratio, AI inoculant supplementation for fermentation, and AI customized product processing).

[0066] The specific testing indicators and results are shown in Table 1 below: Table 1 Based on the three types of indicator data shown in Table 1, the following analysis results were obtained: (1) Fermentation engineering indicators: The organic matter degradation rate of the experimental group was ≥75%, which was 50% higher than that of control group 1 and 25% higher than that of control group 2; the fermentation cycle was shortened to 10-15 days, which was 50% shorter than that of control group 1 and 33% shorter than that of control group 2. This shows that the dosage-based supplementation mechanism of the present invention significantly improves the decomposition efficiency of microorganisms on organic matter, especially the targeted supplementation of microbial agents during the heating and high temperature periods, which effectively maintains the metabolic activity of the microbial community.

[0067] (2) Plant nutrition indicators: The fertilizer nutrient utilization rate of the experimental group was ≥60%, which was 100% higher than that of control group 1 and 50% higher than that of control group 2; the crop seedling biomass was ≥35 g / plant, which was 75% higher than that of control group 1 and 17% higher than that of control group 2. This is due to the product customization module of the present invention: the AI ​​model adds zinc and boron elements that are lacking in the target farmland to the solid fertilizer according to the soil-crop demand matching model, and enriches the root-promoting amino acids and small molecule humic acid in the water-soluble fertilizer, which significantly improves the agronomic effectiveness of the fertilizer.

[0068] (3) Environmental indicators: The carbon emission reduction of the experimental group was ≥0.6 t CO2e / t waste, which was 100% higher than that of control group 1 and 20%-50% higher than that of control group 2; the methane emission of fermentation tail gas was ≤100 mg / m³. 3 Compared to control group 1, the emissions were reduced by 80%, and compared to control group 2, the emissions were reduced by 60%-75%. This is attributed to the AI-controlled process of this invention: precise aeration avoids localized anaerobic environments and inhibits methane generation; at the same time, the biochar module converts solid residue into stable carbon, achieving carbon sequestration and further enhancing the system's carbon reduction capacity.

[0069] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. An AI-based system for the fermentation and nutrient recycling of diverse agricultural waste, characterized in that, include: The pretreatment unit is equipped with an identification sensor module and a mixing and blending module. The identification sensor module identifies the types of agricultural waste and detects its material characteristics. The pretreatment unit performs corresponding crushing or separation treatment according to the type of waste and material characteristics. The mixing and blending module mixes the treated materials with auxiliary materials according to the calculated ratio based on the calculation results of the pre-trained AI model, and adjusts the carbon-nitrogen ratio, moisture content and pH value of the mixture. The AI ​​fermentation unit receives the mixed materials and includes fermentation facilities and its equipped heating and insulation system, sensing system and microbial control system. The sensing system monitors the temperature and oxygen content of the material pile. Based on the monitoring data, the AI ​​fermentation unit controls the operation of the heating and insulation system through an AI model to maintain the fermentation temperature, while controlling the turning and aeration of the pile. The microbial control system inoculates and replenishes compound functional microbial agents into the pile according to the instructions of the AI ​​model. The product processing unit receives the solid materials and fermentation broth obtained after fermentation. The product processing unit includes a solid fertilizer preparation module, a water-soluble fertilizer preparation module, and a biochar preparation module. The solid fertilizer preparation module screens the solid materials, adds trace elements, granulates, and dries them to produce solid organic fertilizer. The water-soluble fertilizer preparation module filters the fermentation broth, uses membrane separation to enrich functional components, and concentrates it to produce organic water-soluble fertilizer. The biochar preparation module pyrolyzes and carbonizes the solid residue to produce biochar and recovers the pyrolysis gas generated during the pyrolysis process.

2. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 1, characterized in that, It also includes an AI monitoring and accounting and digital twin unit, which includes a gas monitoring module, a carbon accounting module, and a digital twin platform; the gas monitoring module monitors greenhouse gas emissions and process energy consumption data generated throughout the entire operation of the system; the carbon accounting module calculates the carbon emission reduction of the system throughout its entire life cycle based on the data obtained by the gas monitoring module and through AI algorithms. The digital twin platform constructs a digital twin model synchronized with the physical system, integrates data from the entire system process to simulate and optimize process parameters, and connects farmland soil data and crop growth data to form a closed-loop control.

3. The AI-based agricultural multi-waste fermentation nutrient recycling system according to claim 2, characterized in that, The gas monitoring module includes gas monitors installed at the inlet and outlet of the fermentation facility, above the pile, and at the tail gas emission outlet of the pyrolysis carbonization furnace, for real-time monitoring of the emission concentrations of methane, carbon dioxide, and nitrous oxide; the gas monitoring module also includes energy consumption monitoring devices installed on waste transport vehicles and each processing unit of the system, for recording transportation fuel consumption and power consumption during the processing. The carbon accounting module is configured to: calculate the system’s carbon emission reduction based on data obtained from the gas monitoring module and energy consumption monitoring device, combined with the CCER methodology for centralized treatment of agricultural waste, using the material balance method and emission factor method. The digital twin platform is built on the Unity3D engine. Its digital twin model is updated synchronously with the physical system at a set frequency. The digital twin platform integrates modules for process parameter optimization, energy consumption monitoring, carbon emission reduction accounting, and product quality traceability. It also accesses data from farmland soil sensors and crop growth monitoring equipment. Based on the accessed data, it uses AI models to simulate and optimize process parameters and adjust the formula and application rate recommendations for fermentation products.

4. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 2, characterized in that, The carbon accounting module is also configured to dynamically predict the value of carbon assets and provide trading recommendations, specifically including: The carbon accounting module connects to an external carbon market data platform to obtain real-time information on carbon quota prices and policy changes. Based on historical and real-time data from the gas monitoring module and energy consumption monitoring device, it uses AI algorithms to predict carbon emission reductions within a set future period. The carbon accounting module takes the predicted carbon emission reductions, real-time carbon quota prices, and policy change information as inputs and uses a built-in value assessment model to calculate and generate a future carbon asset value fluctuation curve. When an extreme point that matches the preset trading strategy appears in the calculated carbon asset value fluctuation curve, the carbon accounting module generates a carbon asset trading opportunity suggestion that includes suggested trading time and suggested trading volume, and outputs the suggestion to the digital twin platform for display and early warning.

5. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 1, characterized in that, The identification sensing module in the pretreatment unit includes an AI visual recognition device and a multi-parameter sensor array, used to identify the type of agricultural waste and detect its moisture content and carbon-nitrogen ratio; The pretreatment unit performs the following treatments based on the identified waste type: for straw waste, it is graded, crushed, and screened according to its lignification degree; for vegetable waste, it is cleaned and chopped; and for livestock and poultry manure, it is separated into solid and liquid components and dried the solid portion. The mixing and blending module includes a mixer, a quantitative feeder, and a device for adding auxiliary materials. Based on the detection data from the identification sensor module, the mixing and blending module calculates the mixing ratio containing auxiliary materials using an AI model. The quantitative feeder transports the processed waste materials and auxiliary materials to the mixer for mixing according to the mixing ratio. During the mixing process, the mixing and blending module adjusts the carbon-nitrogen ratio, moisture content, and pH value of the mixed materials, monitors material parameters in real time, and dynamically adjusts the feeding amount using an AI model.

6. The AI-based agricultural multi-waste fermentation nutrient recycling system according to claim 1, characterized in that, The fermentation facility in the AI ​​fermentation unit is a fermentation shed with an insulated film covering, where the mixed materials are piled up to form a fermentation pile. The heating and insulation system includes solar collectors, phase change heat storage components, and insulation film; the solar collectors are laid on the top of the fermentation shed and connected to the phase change heat storage components through circulation pipes. The thermal insulation film has a multi-layer composite structure; The sensing system includes temperature sensors located at at least two different depths within the stack, and an oxygen sensor located above the stack. The microbial control system includes a storage tank for storing a compound functional microbial agent and a device for applying the agent to the stack. The compound functional microbial agent includes Bacillus subtilis, phosphate-solubilizing bacteria, and nitrogen-fixing bacteria. The AI ​​model is configured to: dynamically control the thermal cycle between the solar thermal collector and the phase change thermal storage components, the opening and closing of the insulation film, the frequency of turning and throwing the pile body and the aeration intensity based on monitoring data from temperature and oxygen sensors; and, based on the assessment of the activity of the microbial community in the pile body, control the microbial regulation system to supplement the pile body with compound functional microbial agents.

7. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 6, characterized in that, The specific method for supplementing compound functional microbial agents in a microbial regulation system controlled by an AI model is as follows: The AI ​​model receives core depth temperature data from the sensor system and cellulase and urease activity data from real-time monitoring of the pile samples. Based on the core depth temperature data, the fermentation process is divided into a warming phase, a high-temperature sustained phase, and a cooling phase. The warming phase is defined as the stage where the core depth temperature rises continuously from the ambient temperature to 55°C; the high-temperature sustained phase is defined as the stage where the core depth temperature remains between 55°C and 60°C; and the cooling phase is defined as the stage where the core depth temperature drops continuously from 55°C to 45°C. During the warming phase, when the cellulase activity falls below a first threshold, a supplementation operation is initiated. The supplementation amount M1 is calculated using the formula M1 = k1 × (A1 - C1), where... k1 is the bacterial replenishment coefficient during the warming period, A1 is the first threshold for cellulase activity, and C1 is the real-time detected cellulase activity value. During the sustained high-temperature period, when the urease activity is lower than the second threshold, the replenishment operation is initiated. The replenishment amount M2 is calculated and determined by the formula M2=k2×(A2-C2), where k2 is the bacterial replenishment coefficient during the high-temperature period, A2 is the second threshold for urease activity, and C2 is the real-time detected urease activity value. The replenished bacterial agents are: during the warming period, mainly the bacterial agent components with cellulose decomposition function in the compound functional bacterial agent; during the sustained high-temperature period, mainly the bacterial agent components with nitrogen conversion function in the compound functional bacterial agent. During the cooling period, when the temperature drops to 45℃ and the activities of both enzymes reach and remain above their corresponding thresholds, the bacterial agent replenishment is stopped.

8. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 1, characterized in that, The solid fertilizer preparation module includes a screening device for screening fermented solid products, an addition device for adding trace elements, a granulator, and a drying device. The solid fertilizer preparation module is configured to: screen the fermented solid products into coarse and fine materials, add trace elements to the fine materials and mix them evenly, granulate the mixed materials, and dry the granulated granules. The water-soluble fertilizer preparation module includes a filtration device, a membrane separation device, a concentration device, and a filling device; The water-soluble fertilizer preparation module is configured to: filter the fermentation liquid product to remove suspended impurities, perform membrane separation on the filtered liquid to enrich functional components, concentrate the enriched liquid, and fill the concentrated liquid. The biochar preparation module includes a pyrolysis carbonization furnace, a cooling device, and a gas processing device; The biochar preparation module is configured to: transport solid residue to a pyrolysis carbonization furnace for anaerobic pyrolysis to generate biochar and pyrolysis gas; cool and screen the generated biochar; purify the pyrolysis gas; and use the purified gas for power generation or directly provide heat energy to the system.

9. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 8, characterized in that, In the solid fertilizer preparation module, the types and proportions of added micronutrients, as well as in the water-soluble fertilizer preparation module, the target molecular weight cutoff range and final solid content of the concentrate in the membrane separation device are determined by an AI model based on soil type data of the target farmland and variety requirements data of the target crop. Soil type data includes soil pH, main nutrient content, and background values ​​of micronutrients; crop variety requirements data includes crop type, growth stage, and historical yield and quality data. The AI ​​model establishes a soil-crop nutrient requirement matching model by analyzing the soil type data and crop variety requirements data. This model first calculates the difference between soil nutrient supply and crop stage requirements, then generates a micronutrient addition formula for the solid fertilizer based on this difference. Simultaneously, it generates a functional component enrichment target for the water-soluble fertilizer based on crop quality improvement requirements. The addition device performs the addition operation according to this micronutrient addition formula, and the membrane separation device in the water-soluble fertilizer preparation module adjusts its operating parameters according to the functional component enrichment target.

10. The AI-based agricultural multi-element waste fermentation nutrient recycling system according to claim 1, characterized in that, The construction and operation of an AI model includes three sequentially executed stages: data fusion, prediction optimization, and instruction generation. The specific construction and operation methods are as follows: During the data fusion phase, the AI ​​model simultaneously receives and integrates multi-source time-series data, including waste type, moisture content and carbon-nitrogen ratio data from the identification sensor module, multi-depth temperature data and oxygen content data of the pile from the AI ​​fermentation unit sensor system, and online detection data of solid product screening efficiency, liquid product functional component concentration and biochar yield from the product processing unit. During the prediction and optimization phase, the core processor of the AI ​​model calls a multi-objective prediction model trained on historical fermentation data. The multi-objective prediction model takes real-time time-series data obtained in the data fusion phase as input and outputs predicted values ​​for the core temperature trend of the pile, the activity trend of key microbial enzymes, and the yield trend of the target product within a set future time period. The AI ​​model compares the predicted values ​​with the preset process target values ​​for each stage and, based on the comparison results, uses an optimization algorithm to solve for a set of controllable process parameter combinations that make the predicted values ​​closest to the process target values. The controllable process parameter combinations include mixing ratio, turning frequency, aeration intensity, microbial agent replenishment amount, and key operating parameter settings for the product processing module. During the instruction generation stage, the AI ​​model converts the controllable process parameter combination obtained in the prediction and optimization stage into a specific sequence of control instructions with time-series markers, and sends them to the actuators in the mixing and blending module, the AI ​​fermentation unit, and the various preparation devices in the product processing module. The AI ​​model is also equipped with a model update mechanism: it associates the final product quality indicators, actual energy consumption data and multi-source time series data of the whole process in each complete fermentation batch to form a new data package, and periodically uses the newly accumulated data package to retrain the multi-objective prediction model to update its model parameters.

Images