Fucoxanthin synthesis optimization method and system based on reinforcement learning

Abstract

The invention relates to the technical field of biological environment engineering, and provides a fucoxanthin synthesis optimization method and system based on reinforcement learning, and the method comprises the following steps: a real-time data acquisition system; constructing a state space; a deep reinforcement learning decision network; optimizing a dynamic strategy; and strategy verification is linked with the process. Through a light quality-stimulant collaborative reinforcement learning regulation strategy, the static operation limitation of a traditional process is broken through, the photosynthetic physiological state of microalgae is perceived in real time, the light environment and chemical signals are dynamically optimized, the fucoxanthin synthesis efficiency is remarkably improved, and a multi-omics closed-loop verification mechanism is combined, so that the fucoxanthin synthesis efficiency is improved. The molecular cascade effect of a light signal driving pigment synthesis pathway is illuminated, and a precise target spot is provided for metabolic engineering transformation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological environment engineering, and particularly relates to a fucoxanthin synthesis optimization method and system based on reinforcement learning. BACKGROUND

[0002] Fucoxanthin is a photosynthetic pigment synthesized by diatom cells, which has anti-inflammatory, antioxidant, antitumor and anti-obesity activities, and has a wide application prospect in the field of medicine and health care. The traditional production method relies on large-scale seaweed extraction, which has defects such as low pigment content, long cultivation cycle, and easy influence by environmental pollutants, resulting in high raw material cost and insufficient product purity. Although microalgae cultivation technology can overcome some limitations, the existing process still relies on the experience trial-and-error method in regulating fucoxanthin synthesis efficiency. The light quality regulation is mostly limited to fixed waveband ratio, and lacks real-time adaptation to photosynthetic physiological dynamic response. The exogenous stimulant addition strategy is single, and cannot be linked with cell metabolic state, resulting in unstable product accumulation. More importantly, the regulation mechanism of fucoxanthin biosynthesis pathway has not been systematically analyzed, especially lacking cross-scale correlation research on light signal and metabolite accumulation, which makes the process optimization lack theoretical support. In addition, the existing extraction process relies on toxic organic solvents and has complicated purification steps, which is difficult to meet the requirements of green production.

[0003] These technical shortcomings seriously restrict the application of fucoxanthin, so it is necessary to develop a fucoxanthin synthesis optimization method and system based on reinforcement learning. SUMMARY

[0004] In order to overcome the general shortcomings of the prior art, the present application provides a fucoxanthin synthesis optimization method and system based on reinforcement learning.

[0005] The technical scheme of the present application is: a fucoxanthin synthesis optimization method based on reinforcement learning, comprising the following steps: Step 1: Real-time data acquisition system, integrating photosynthetic physiological monitoring equipment, biomass sensor and metabolite analyzer, synchronously acquiring light system parameters, biological macromolecule conversion spectrum and fucoxanthin content of the golden odontella cultivation system; Step 2: State space construction, fusing environmental parameters, physiological response indexes and metabolite data into a 23-dimensional state vector; Step 3: Deep reinforcement learning decision network, adopting Actor-Critic framework, Actor network outputting light quality formula and dynamic addition amount of exogenous stimulant, Critic network evaluating state value function; Step 4: Dynamic strategy optimization, taking the increase amplitude of fucoxanthin yield and the biomass production per unit energy consumption as the joint reward function, updating network weight through PPO algorithm, and generating the optimal regulation strategy; Step 5: Strategy verification and process linkage, transmit the optimization instructions to the LED light source controller and peristaltic pump actuator, and trigger the green purification process to extract the product.

[0006] As a preferred embodiment of the present application, the light quality regulation strategy specifically comprises: decomposing white light into three independently controllable wavebands of blue light, red light and far-red light, and stimulating the expression of key enzymes in the fucoxanthin synthesis pathway by adjusting the intensity ratio of the three wavebands.

[0007] As a preferred embodiment of the present application, the exogenous stimulant response analysis is realized by constructing an FTIR-PAM data fusion model: the change in the intensity of the characteristic peak of the ester bond and amide I band detected by FTIR represents the lipid-protein conversion rate, and the non-photochemical quenching parameters measured by PAM are combined to predict the accumulation of fucoxanthin under the stimulation of triacontanol.

[0008] As a preferred embodiment of the present application, the strategy verification mechanism includes transcriptome / metabolome closed-loop verification: when the reinforcement learning strategy triggers the peak value of fucoxanthin synthesis rate, the algal cell sample is automatically collected for RNA-Seq analysis to screen the expression changes of key genes in pigment synthesis, and the concentration of fucoxanthin precursor in the metabolome is correlated to form a molecular mechanism verification report.

[0009] As a preferred embodiment of the present application, the green purification process linkage specifically comprises: when the intracellular concentration of fucoxanthin detected by HPLC is ≥6mg / g DCW, the low-temperature extraction of ethanol / water mixed solvent is automatically started, followed by primary filtration through a ceramic membrane with a pore size of 0.22μm, and then gradient purification is performed using a reverse phase C18 chromatographic column, and the final product purity is ≥80%.

[0010] As a preferred embodiment of the present application, the dynamic strategy optimization module is embedded with multi-objective constraint conditions: the light energy utilization rate and solvent toxicity index are used as hard constraints to ensure that the output strategy of the algorithm meets the green preparation standard.

[0011] As a preferred embodiment of the present application, the training data of the reinforcement learning model is derived from a digital twin simulation system: based on multi-omics correlation analysis, a fucoxanthin metabolic pathway dynamics model is constructed to simulate the carbon flow distribution path under different light quality-stimulant combinations, and virtual training samples are generated to accelerate the convergence of the strategy.

[0012] A fucoxanthin synthesis optimization system based on reinforcement learning, comprising an environment regulation unit, a physiological monitoring unit, an edge computing server and a purification execution unit, realizes the whole-process closed-loop control from algal culture to high-purity fucoxanthin.

[0013] By adopting the above technical solution, the present application has the following advantages: 1. The application breaks through the static operation limitation of traditional process through light quality-stimulant synergistic reinforcement learning regulation strategy, realizes real-time perception of microalgae photosynthetic physiological state and dynamic optimization of light environment and chemical signal, significantly improves the synthesis efficiency of fucoxanthin, and combines multi-omics closed loop verification mechanism to clarify the molecular cascade effect of light signal driven pigment synthesis pathway, and provides precise target for metabolic engineering modification.

[0014] 2. The application realizes linkage of intelligent decision module, low-toxicity solvent extraction and membrane chromatography purification process, significantly reduces environmental load while ensuring high purity of product, and realizes closed loop automation of production process, which significantly reduces manual intervention, and sets a new paradigm for low-cost and sustainable industrialization of fucoxanthin. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The step flow chart of the synthesis optimization method of the application.

[0016] Figure 2 The structure schematic diagram of the system of the application. DETAILED DESCRIPTION

[0017] In this text, the specific features, structures or characteristics described in conjunction with the embodiments can be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it independent or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0018] The system regulates the culture environment of Chrysophyte Odontella through a multi-spectral LED array (450nm blue light / 650nm red light / 700nm far red light dynamic ratio, intensity ratio 5:3:2→5:2:3) and an exogenous stimulant precise dosing module (triacontanol concentration 0.1-10μM), and synchronously collects photosynthetic parameters such as Fv / Fm, rETR and qP by an online monitoring unit (Xe-PAM); analyzes 1740cm -1 ester bond / 1650cm -1Amide I band intensity; HPLC to detect fucoxanthin content) to construct a 23-dimensional state vector; based on the deep reinforcement learning Actor-Critic framework, the increase in fucoxanthin yield (target > 10%) and the biomass production per unit of light energy as the joint reward function, dynamically optimize the light quality formula and the dose of stimulants; when the intracellular concentration of fucoxanthin is ≥6 mg / g DCW, automatically trigger the green purification process: using ethanol / water (7:3 v / v) mixed solvent 40℃ extraction→0.22μm ceramic membrane primary filtration→C18 reverse phase chromatographic column (methanol / acetic acid water = 85:15 gradient elution), to ensure that the product purity is ≥80%; At the same time, through the multi-omics closed-loop verification mechanism (RNA-Seq analysis of ZDS / CRTISO / LYCB gene expression up-regulation fold, metabolomics detection of fucoxanthol accumulation rate) to analyze the molecular mechanism of the regulation strategy (such as when the red light ratio >60%, the expression of pigment synthesis genes is increased by 2.3 times), and finally form an intelligent control system from parameter optimization, mechanism verification to high-purity product preparation.

[0019] Specifically, a fucoxanthin synthesis optimization method based on reinforcement learning, as shown in Figure 1 includes the following steps: Step 1: Real-time data acquisition system, integrated photosynthetic physiological monitoring equipment, biomass sensor and metabolite analyzer, synchronous acquisition of gold Odu algae culture system light system parameters, biological macromolecule conversion spectrum and fucoxanthin content; Step 2: State space construction, environmental parameters, physiological response indicators, and metabolic product data are fused into a 23-dimensional state vector; Step 3: Deep reinforcement learning decision network, using Actor-Critic framework, Actor network outputs light quality formula and dynamic addition amount of exogenous stimulants, Critic network evaluates state value function; Step 4: Dynamic strategy optimization, taking the increase in fucoxanthin yield and the biomass production per unit of energy as the joint reward function, updating the network weight through PPO algorithm, and generating the optimal regulation strategy; Step 5: Strategy verification and process linkage, the optimized instructions are transmitted to the LED light source controller and peristaltic pump actuator, and the green purification process is triggered to extract the product.

[0020] Based on the above steps, the specific content is as follows: The light quality regulation strategy is as follows: white light is decomposed into blue light (450±10nm), red light (650±10nm) and far red light (700±10nm) three independently controllable wavebands, by adjusting the intensity ratio of the three wavebands (blue light: red light: far red light = 5:3:2→5:2:3 dynamic range), to stimulate the expression of key enzymes in the fucoxanthin synthesis pathway; Exogenous stimuli response analysis is realized by constructing FTIR-PAM data fusion model: using FTIR to detect the intensity changes of characteristic peaks at 1740 cm -1 (estercoupling), 1650 cm -1 (amide I band) to represent the conversion rate of lipid-protein, combined with the non-photochemical quenching (NPQ) parameter measured by PAM, to predict the fucoxanthin accumulation under the stimulation of triacontanol; The strategy verification mechanism includes transcriptome / metabolome closed-loop verification: when the reinforcement learning strategy triggers the peak of fucoxanthin synthesis rate, the algal cell samples are automatically collected for RNA-Seq analysis to screen the expression changes of key genes (such as ZDS, CRTISO, LYCB) of pigment synthesis, and to correlate the concentrations of fucoxanthin precursors (fucoxanthol, diadinoxanthin) in the metabolome, forming a molecular mechanism verification report; The green purification process linkage is as follows: when the intracellular concentration of fucoxanthin is ≥6 mg / g DCW detected by HPLC, the low-temperature (40±2℃) extraction of ethanol / water (7:3 v / v) mixed solvent is automatically started, followed by primary filtration through a ceramic membrane with a pore size of 0.22 μm, and then gradient purification is carried out using a reverse phase C18 chromatographic column (eluent: methanol / acetic acid aqueous solution=85:15), and the final product purity is ≥80%.

[0021] The dynamic strategy optimization module is embedded with multi-objective constraint conditions: taking the light energy utilization rate (μmol photons / m 2 / s / W) and the solvent toxicity index (LD50>5000 mg / kg) as hard constraints to ensure that the output strategy of the algorithm meets the green preparation standards; The training data of the reinforcement learning model comes from the digital twin simulation system: based on multi-omics correlation analysis, a fucoxanthin metabolic pathway dynamics model is constructed to simulate the carbon flow distribution path under different light quality-stimulant combinations, generating a virtual training sample to accelerate the convergence of the strategy.

[0022] A fucoxanthin synthesis optimization system based on reinforcement learning, as Figure 2 shown, includes an environmental regulation unit, a physiological monitoring unit, an edge computing server, and a purification execution unit, realizing the whole-process closed-loop control from algae cultivation to high-purity fucoxanthin.

[0023] Specifically, the environmental regulation unit uses a cylindrical glass reactor (diameter 30 cm, height 50 cm, effective volume 30 L) with multiple layers of LED light source arrays installed inside, including three independently controlled wavebands: blue light (center wavelength 450±2 nm, intensity range 0-300 μmol / m 2Red light (650 ± 2 nm, equal intensity range), far-red light (700 ± 2 nm, equal intensity range), temperature control maintained at 25 ± 0.5℃ by a sandwich water circulation system, pH sensor real-time monitoring and automatic adjustment to 7.8 ± 0.2, exogenous stimulant dosing module configured with triacontanol storage tank (1 mM ethanol solution), injected into the reactor by high-precision peristaltic pump (flow error <0.05 mL / min) as needed, concentration control range 0.1-10 μM; In the physiological monitoring unit, the Xe-PAM chlorophyll fluorescence instrument was scanned every 2 hours to measure the maximum quantum yield of photosystem II Fv / Fm (range 0.65-0.85, accuracy ± 0.03), relative electron transport rate rETR (range 0-200 μmole electrons / m 2 2 / s), photochemical quenching qP (range 0-1), and biological macromolecule analysis was performed using a Fourier transform infrared spectrometer (FTIR) to sample and detect every 12 hours, with ester bond vibration peaks at 1740 cm -1 -1 (characterizing lipid accumulation) and amide I band at 1650 cm -1 -1 (characterizing protein transformation), wave number resolution 4 cm -1 -1, high-performance liquid chromatography (HPLC) system equipped with a C18 reversed-phase chromatographic column (4.6x250 mm, 5 μm particle size), mobile phase methanol-water (85:15 v / v), flow rate 1.0 mL / min, fucoxanthin detected at 450 nm wavelength (retention time 12.3 ± 0.2 min), detection limit 0.01 mg / g DCW.

[0024] The edge computing server is equipped with an NVIDIA Jetson AGX Xavier module, the operating system is Ubuntu 20.04, the PyTorch 1.10 framework under Python 3.8 environment is deployed, and the control instruction output resolution is: light intensity adjustment step 10 μmol / m 2 2 / s, triacontanol concentration adjustment step 0.1 μM.

[0025] It should be noted that the reinforcement learning model construction and training process is as follows: State space definition, the following 23-dimensional parameters are normalized to the interval [0, 1] to form a state vector: Environmental parameters (6 dimensions): blue light intensity, red light intensity, far-red light intensity, triacontanol concentration, temperature, pH value; Physiological parameters (9 dimensions): biomass (OD680 value), specific growth rate (μ, h -1 ), Fv / Fm, rETR, qP, FTIR 1740 cm -1Peak area, FTIR 1650 cm -1 Peak area, chlorophyll a / c ratio Metabolic parameters (8 dimensions): fucoxanthin content (mg / g DCW), precursor phlorofucohydrol concentration (ng / mL), distribution ratio of carbon flux towards pigment synthesis (%).

[0026] Decision network architecture, Actor network: 3-layer fully connected neural network (128-64-32 nodes), input layer receives 23-dimensional state vector, output layer generates 4-dimensional action vector (blue light intensity, red light intensity, far-red light intensity, triacontanol concentration), activation function output layer uses Sigmoid to constrain the action value range; Critic network: double-flow structure to evaluate state value, the first branch evaluates the immediate reward, the second branch predicts long-term income, and the loss function uses Huber Loss.

[0027] Reward function design: ; Wherein, is the initial fucoxanthin yield (default 1.5 mg / L / day), is the cumulative light energy input at time t, is the solvent toxicity index (penalty term +0.5 when ethanol concentration >10%).

[0028] Whole-process collaborative control and verification mechanism: Light quality-stimulant dynamic regulation: when the Critic network evaluates that increasing red light intensity can increase income, output a formula with red light ratio >60% (example: blue light: red light: far-red light = 5:3:2 → adjust to 5:4:1), and simultaneously trigger triacontanol pulse addition (2 μM for 2 hours), then monitor the physiological response, and FTIR detects that when the peak area of 1740 cm -1 decreases by 20% and the peak area of 1650 cm -1 increases by 15%, it is determined that the carbon flow towards protein synthesis is shifted, and the far-red light ratio is automatically reduced by 5%; Multisomic verification closed loop: when the fucoxanthin synthesis rate is >0.8 mg / L / h for 6 consecutive hours, automatically collect 50 mL of algal liquid for omics analysis: Transcriptome: RNA-Seq detects the expression amount of pigment synthesis pathway genes (ZDS, CRTISO, LYCB), and the expression amount of LYCB gene increases by 2.3 times under red light stimulation (qPCR verification p<0.01); Metabolome: LC-MS / MS targeted quantification of precursor phlorofucohydrol, with a concentration peak of 35 ng / mL (↑40% compared to baseline).

[0029] Green purification process linkage: The trigger condition is that the intracellular concentration of fucoxanthin detected by HPLC is ≥6 mg / g DCW, the extraction process uses algal sludge mixed with ethanol / water (7:3 v / v), ultrasonic extraction at 40±2℃ for 30 min (power 300 W), and the supernatant is collected by centrifugation. The specific process is as follows: First step: 0.22 μm ceramic membrane filtration, molecular weight cutoff >100 kDa impurities; Second step: C18 reverse phase column (Φ50×250 mm), methanol-0.1% acetic acid aqueous solution (85:15 v / v) gradient elution, collection of 12-13 min fraction; Third step: HPLC-UV detection purity ≥82.5% (n=3), solvent residue <10 ppm.

[0030] In summary, after the system is started, the 30L cylindrical photobioreactor is first initialized: inject the modified L1 medium (sodium nitrate concentration 875mg / L, disodium hydrogen phosphate 36.8mg / L) sterilized by 0.22μm filter membrane, inoculate Chlorella variabilis (initial OD680=0.1), set temperature 25±0.5℃, pH 7.8±0.2, and pass in mixed gas containing 1% CO2 (flow rate 0.1vvm); turn on the basic white light mode of the multi-spectrum LED array (blue light 450nm / red light 650nm / far red light 700nm, intensity ratio 5:3:2, total light intensity 150μmol / m 2 / s), the online monitoring module is activated simultaneously - Xe-PAM chlorophyll fluorescence instrument scans Fv / Fm (initial value 0.75±0.03), rETR (initial value 85μmol electrons / m 2 / s), qP (initial value 0.68) every 2 hours, FTIR spectrometer collects algal body macromolecular spectrum (1740cm -1 ester bond initial peak area 1200±50, 1650cm -1 amide I band initial peak area 950±40) every 12 hours, and HPLC system detects fucoxanthin content (baseline value 1.2mg / g DCW) every day; At this time, the edge computing server loads the pre-trained reinforcement learning model (metabolic pathway simulation model generated based on 50,000 groups of digital twin data), and the initial environmental parameters (three-band light intensity, temperature, pH), physiological parameters (biomass growth rate 0.15h -1, Fv / Fm values, FTIR feature peak intensity ratios) and metabolic parameters (fucoxanthin synthesis rate 0.2 mg / L / h) into a 23-dimensional state vector (normalized to the [0, 1] interval) input to the Actor-Critic network; the Actor network (3-layer fully connected structure, 128-64-32 nodes) outputs the first control instruction according to the instant reward value (benchmark reward function R = 0.7 x Δ fucoxanthin yield + 0.3 x unit light biomass - 0.1 x solvent toxicity) evaluated by the Critic network: increase red light intensity to 180 μmol / m 2 / s (60% of total light intensity), reduce far-red light to 45 μmol / m 2 / s (15% of the total), and simultaneously start the triacontanol pulse addition (creep pump injection of 1 mM stock solution, final concentration 2 μM for 2 hours); at the 6th hour after the execution of the instruction, Xe-PAM detects that the rETR value rises to 120 μmol electrons / m 2 / s (an increase of 41%), and FTIR shows that the 1740 cm -1 peak area decreases to 980 (a decrease of 18%), and the 1650 cm -1 peak area increases to 1120 (an increase of 18%), indicating that photosynthetic carbon flow is diverted from lipid synthesis to protein accumulation; the Critic network calculates the reward value R to rise by 0.25 based on the new state vector (rETR > 120 μmol electrons / m 2 / s, 1740 / 1650 cm -1 peak area ratio <0.9), triggering policy optimization - the Actor network outputs the second control instruction to reduce the proportion of far-red light to 10% (three-band ratio 5:4:1); after 24 hours of continuous control, HPLC detects that the fucoxanthin concentration reaches 3.8 mg / g DCW (an increase of 217% compared with the baseline), and the biomass OD680=0.45 (specific growth rate 0.18 h -1), at this time the system determines that the rate of fucoxanthin synthesis is >0.8 mg / L / h for 6 consecutive hours (meeting the omics sampling threshold), and 50 mL of algal liquid is automatically collected to start the multi-omics verification closed loop: RNA-Seq analysis shows that under the red light dominant strategy, the expression of the key gene LYCB in the pigment synthesis pathway is up-regulated by 2.3 times (p=0.003), and the ZDS gene is up-regulated by 1.8 times (p=0.007), and the LC-MS / MS metabolic group detects the precursor phaeophorbide concentration of 35 ng / mL (up by 40% compared with the baseline), confirming that light quality regulation activates the carotenoid synthesis pathway to improve fucoxanthin accumulation; when the intracellular concentration of fucoxanthin breaks through the threshold of 6 mg / g DCW, the system links the green purification process - the algal mud is collected by centrifugation at 4000 rpm for 10 minutes, mixed with an ethanol / water (7:3 v / v) mixture at a ratio of 1:5 (w / v) at 40±2°C, and ultrasonic extracted for 30 minutes (power 300W), and the centrifugal supernatant is filtered through a 0.22 μm ceramic membrane (retention >100 kDa impurity protein), and the filtrate is injected into a C18 reverse phase chromatography column (Φ50×250mm, filler particle size 5 μm), eluted with a methanol-0.1% acetic acid aqueous solution (85:15 v / v) gradient (flow rate 2 mL / min), and the main peak fraction of fucoxanthin at 12.3±0.2 min was collected; the final product was verified by HPLC-ELSD to have a purity of 83.2±1.1% (n=3), and the solvent residue was <10 ppm, meeting the green chemistry standard (LD50>5000 mg / kg); the whole process ran for 72 hours, with a cumulative yield of 18.7 mg / L, an increase of 15.4% over the traditional process, and a unit light energy biomass output of 0.33 g / μmol photons, and the system automatically generated a regulation report (including light quality formula optimization path, gene expression heat map, and purification recovery rate curve), completing the whole closed loop production from algal culture to intelligent regulation to molecular mechanism analysis to high-purity product preparation.

[0031] The above examples are provided to those skilled in the art to implement or use the present application, and those skilled in the art can make various modifications or changes to the above examples without departing from the spirit of the present application, so the protection scope of the present application is not limited by the above examples, but should be the maximum scope meeting the innovative features mentioned in the claims.

Claims

1. A fucoxanthin synthesis optimization method based on reinforcement learning, characterized by, Comprise the following steps: Step 1: Real-time data acquisition system, integrated photosynthetic physiological monitoring equipment, biomass sensor and metabolite analyzer, synchronous acquisition of golden odium culture system of light system parameters, biological macromolecular conversion spectrum and fucoxanthin content; Step 2: State space construction, environmental parameters, physiological response indicators, and metabolic product data are fused into a 23-dimensional state vector; Step 3: Deep reinforcement learning decision network, using Actor-Critic framework, Actor network outputs light quality formula and exogenous stimulant dynamic addition amount, Critic network evaluates state value function; Step 4: Dynamic strategy optimization, taking the increase amplitude of fucoxanthin yield and the biomass production per unit energy consumption as the joint reward function, updating the network weight through PPO algorithm, and generating the optimal regulation strategy; Step 5: Strategy verification and process linkage, the optimized instructions are transmitted to the LED light source controller and peristaltic pump actuator, and the green purification process is triggered at the same time to extract the product.

2. The fucoxanthin synthesis optimization method based on reinforcement learning according to claim 1, characterized in that, The light quality regulation strategy is as follows: white light is decomposed into blue light, red light and far red light, which are independently controllable, and the intensity ratio of the three wavebands is adjusted to stimulate the expression of key enzymes in the fucoxanthin synthesis pathway.

3. The fucoxanthin synthesis optimization method based on reinforcement learning according to claim 1, characterized in that, The exogenous stimulant response analysis is realized by constructing an FTIR-PAM data fusion model: FTIR is used to detect ester bond and amide I band, and the intensity change of characteristic peak is used to represent the lipid-protein conversion rate, combined with the non-photochemical quenching parameters measured by PAM, to predict the accumulation of fucoxanthin under the stimulation of triacontanol.

4. The fucoxanthin synthesis optimization method based on reinforcement learning according to claim 1, characterized in that, The strategy verification mechanism includes transcriptome / metabolome closed-loop verification: when the reinforcement learning strategy triggers the peak value of fucoxanthin synthesis rate, the algal cell sample is automatically collected for RNA-Seq analysis to screen the expression change of key genes in pigment synthesis, and the concentration of fucoxanthin precursor in the metabolome is correlated to form a molecular mechanism verification report.

5. The fucoxanthin synthesis optimization method based on reinforcement learning according to claim 1, characterized in that, The green purification process linkage is as follows: when the intracellular concentration of fucoxanthin is detected by HPLC ≥ 6 mg / g DCW, the low-temperature extraction of ethanol / water mixed solvent is automatically started, followed by primary filtration through ceramic membrane with pore size of 0.22 μm, and then gradient purification is carried out by using reverse phase C18 chromatographic column, and finally the purity of the product is ≥ 80%.

6. The fucoxanthin synthesis optimization method based on reinforcement learning according to claim 1, characterized in that, The dynamic strategy optimization module is embedded with multi-objective constraint conditions: the light energy utilization rate and solvent toxicity index are used as hard constraints to ensure that the algorithm output strategy meets the green preparation standard.

7. The fucoxanthin synthesis optimization method based on reinforcement learning according to claim 1, characterized in that, The training data of the reinforcement learning model comes from the digital twin simulation system: based on multi-omics correlation analysis, a kinetic model of fucoxanthin metabolic pathway is constructed to simulate the carbon flow distribution path under different light quality-stimulant combinations, and virtual training samples are generated to accelerate the convergence of the strategy.

8. A fucoxanthin synthesis optimization system based on reinforcement learning, characterized by, The integrated method of any one of claims 1-7 comprises an environmental regulation unit, a physiological monitoring unit, an edge computing server and a purification execution unit, realizing the whole process closed-loop control from algal culture to high-purity fucoxanthin.

Images