Haematococcus cultivation dissolved oxygen coupling regulation method for corrosion-resistant photoreactor

Abstract

The application discloses a haematococcus cultivation dissolved oxygen coupling regulation method for a corrosion-resistant photoreactor, relates to the field of bioengineering, and builds a photoreactor and a matching regulation system made of a corrosion-resistant material; a special improved BG-11 culture medium for haematococcus is prepared and is injected into the photoreactor; a dynamic method is used to calibrate the oxygen transfer coefficient under different aeration rates, a ternary equation of light intensity, cell density and oxygen production rate is fitted in combination with the photosynthetic characteristics of haematococcus, and a dissolved oxygen dynamic prediction basic model is established. The photoreactor and the matching regulation system made of the corrosion-resistant material are built, the interference of material dissolution on the culture medium is effectively avoided, and the stability of the haematococcus cultivation environment is ensured. The oxygen transfer coefficient calibrated based on the dynamic method is combined with the ternary fitting equation of light intensity, cell density and oxygen production rate, a dissolved oxygen dynamic prediction basic model is constructed, and model prediction control and reinforcement learning intelligent agents are combined, so that precise coupling regulation of dissolved oxygen is realized.

Description

Technical Field

[0001] This invention relates to the field of bioengineering technology, and in particular to a method for coupled regulation of dissolved oxygen in the cultivation of *Rhodotorula pulmonata* for use in corrosion-resistant photoreactors. Background Technology

[0002] As the core astaxanthin-producing strain, *Hylococcus pluvialis* relies on a controlled photosynthetic environment constructed using a photoreactor (PBR) for its industrial cultivation. Dissolved oxygen (DO) concentration is a key regulatory factor throughout the entire cultivation cycle of *Hylococcus pluvialis*: during the biomass accumulation stage, DO needs to be maintained within a suitable photosynthetic range to ensure rapid cell division; during the astaxanthin synthesis stage, fluctuations in DO directly inhibit the activity of carotenoid synthases, leading to a sharp drop in product yield. However, current dissolved oxygen regulation technology for *Hylococcus pluvialis* cultivation still faces multiple technical bottlenecks, severely restricting the stability and economic viability of industrial production.

[0003] Firstly, the materials used in the photoreactor and its components are not sufficiently compatible. The culture medium commonly used for *Hylococcus pluvialis* is weakly acidic (pH 7.0-8.5) and contains high concentrations of inorganic salts (such as NaHCO3 and MgSO4). Traditional photoreactors often use ordinary borosilicate glass or carbon steel, which are prone to corrosion during long-term use. Simultaneously, if existing gas mass flow controllers (MFCs) are made of ordinary stainless steel, their inner walls are easily corroded by culture medium vapors, forming rust spots. This leads to a decrease in the accuracy of aeration control, further exacerbating displacement (DO) fluctuations.

[0004] Secondly, dissolved oxygen regulation lacks a precise dynamic prediction model. Changes in DO concentration are the result of the synergistic effect of the photosynthetic oxygen production rate (OPR) of Haematococcus pluvialis and the oxygen transfer rate (OTR) of the reactor: OPR depends on the dynamic coupling of light intensity, cell density and metabolic state, while OTR is related to aeration rate, stirring rate and oxygen transfer coefficient (kLa).

[0005] Third, no differentiated regulatory strategies were designed for the two-stage physiological characteristics of Haematococcus pluvialis.

[0006] Fourth, there is a lack of effective mechanisms to deal with dynamic disturbances during the cultivation process.

[0007] Fifth, there is a lack of synergistic optimization between energy consumption and product efficiency.

[0008] In summary, existing technologies have significant shortcomings in areas such as photoreactor material adaptation, dissolved oxygen dynamic prediction, staged control logic, interference response, and energy consumption optimization. There is an urgent need for a dissolved oxygen coupling control scheme that can adapt to the physiological characteristics of Haematococcus pluvialis and take into account both precise control and cost control, so as to break through the technical bottleneck of industrial cultivation of Haematococcus pluvialis. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a method for coupled dissolved oxygen regulation in the cultivation of *Rhodotorula pulmonata* for corrosion-resistant photoreactors. The technical solution is as follows:

[0010] A method for coupled dissolved oxygen regulation in the cultivation of *Rhodotorula glutinis* for corrosion-resistant photoreactors includes the following steps:

[0011] Step 1: Construct a photoreactor and its control system made of corrosion-resistant materials;

[0012] Step 2: Prepare a modified BG-11 medium for Haematococcus pluvialis and inject it into the photoreactor; calibrate the oxygen transfer coefficient under different aeration rates using a dynamic method, and establish a basic model for dynamic prediction of dissolved oxygen by fitting a ternary equation of light intensity, cell density, and oxygen production rate based on the photosynthetic characteristics of Haematococcus pluvialis.

[0013] Step 3: Introduce the logarithmic-stage Haematococcus pluvialis seed liquid into the photoreactor and simultaneously start the model prediction controller and reinforcement learning agent;

[0014] Step 4: When Haematococcus pluvialis is in the biomass accumulation stage, collect data on dissolved oxygen (DO), light intensity, aeration parameters, biofilm status, and culture medium characteristics. After data preprocessing, transmit the data to the control unit. Periodically use monitoring data to calibrate the dissolved oxygen prediction model, and adjust the light intensity and aeration parameters in a coordinated manner according to preset logic based on DO deviation. Periodically start RL training, update MPC control parameters based on culture performance, and perform biofilm cleaning operations as needed.

[0015] Step 5: When it is determined that Haematococcus pluvialis has reached the preset biomass threshold and entered the astaxanthin synthesis stage, reset the DO control interval, model parameters and reward function of the reinforcement learning agent of the model prediction controller, and simultaneously set the induction period culture conditions; correct the dissolved oxygen dynamic prediction basic model based on the metabolic characteristics of Haematococcus pluvialis during the induction period, and adjust the DO regulation priority; start RL training at high frequency to dynamically optimize the DO control interval and reward function weights.

[0016] Optionally, post-culture optimization steps are also included: when the astaxanthin content is determined to be greater than or equal to the cell dry weight threshold, the DO control accuracy, product yield and energy consumption indicators of this culture are evaluated after the culture is completed, and the culture data is added to the dissolved oxygen prediction model library and the reinforcement learning agent experience library; global RL training is carried out periodically to optimize the general parameters of cross-batch regulation and realize the self-iteration of the regulation strategy.

[0017] Optionally, in step 1, the supporting control system includes a corrosion-resistant DO electrode, a gas mass flow controller, a quartz-encased photosensitive sensor, and an ultrasonic biofilm monitoring sensor; the corrosion-resistant material is made of quartz glass or polyvinylidene fluoride; the gas mass flow controller is made of 316L stainless steel; the outer shell of the corrosion-resistant DO electrode is made of titanium alloy and the sealing element is made of fluororubber; the thickness of the light-transmitting layer of the quartz-encased photosensitive sensor is greater than or equal to 2mm, ensuring compatibility with the weakly acidic or high-salt culture medium environment of Haematococcus pluvialis culture and eliminating the risk of material leaching.

[0018] Optionally, the composition of the modified BG-11 medium specifically for *Rhodotorula pulmonale* in step 2 is as follows:

[0019] The culture medium is prepared with 1.2 g / L to 1.8 g / L NaNO3, 0.4 g / L to 0.6 g / L NaHCO3, and 0.06 g / L to 0.09 g / L MgSO4·7H2O. The pH of the culture medium is adjusted to 7.5 to 8.5. Sterilization is achieved by filtration using a polyethersulfone membrane, and a corrosion-resistant PE storage tank is used during the preparation process.

[0020] Optionally, in step 2, the dynamic method for calibrating the oxygen transfer coefficient specifically involves: introducing N2 into the photoreactor under algae-free conditions to reduce DO to below 10%, then switching to air, recording the DO recovery curve, and fitting the aeration rate kLa correlation equation:

[0021] kLa = 0.04 - 0.06 × Gas_set + 0.01 - 0.03; where Gas_set is the aeration rate setting; representing the volume of gas introduced per minute per unit volume of culture medium;

[0022] The ternary equation fitting the light intensity, cell density, and oxygen production rate is as follows:

[0023] ;

[0024] in This is the actual oxygen production rate of Haematococcus pluvialis. This is the maximum oxygen production rate of Haematococcus pluvialis. It is the initial light intensity of the photoreactor. is the light intensity half-saturation constant, k is the light absorption coefficient of *Rhodotorula purpureus* cells, C is the *Rhodotorula purpureus* cell density, and L is the optical path length of the photoreactor. It is the cell density half-saturation constant. It is the system correction coefficient. It is a nutrient limitation correction factor, determined by measuring the nitrogen concentration in the culture medium and substituting it into the empirical formula: ,in The concentration of inorganic nitrogen that can be directly utilized by Haematococcus pluvialis in the culture medium. These are characteristic parameters of nitrogen nutrition.

[0025] Optionally, the preset logic in step 4 is: when DO is higher than the upper limit of the control range for the biomass accumulation stage, use 40-60 μmol / m 2 If the light intensity is reduced by an increment of 0.04-0.06 vvm, and the DO is still found to be above the limit, the ventilation rate is reduced by an increment of 0.04-0.06 vvm. If the DO is below the lower limit of the control range, the ventilation rate is increased by an increment of 0.04-0.06 vvm. If the DO is still not up to standard, the ventilation rate is increased by an increment of 40-60 μmol / m. 2 / s amplitude increases light intensity.

[0026] Optionally, the biofilm cleaning operation is as follows: when the ultrasonic biofilm monitoring sensor detects a membrane thickness ≥50μm, first purge with 0.08-0.12MPa sterile compressed air for 10-15min. If the DO measurement deviation is still ≥10%, then inject 0.08%-0.12% food-grade citric acid solution for circulation and rinsing for 5-8min. Finally, replace with sterile culture medium.

[0027] Optionally, the induction period culture conditions in step 5 include: temperature 26-30℃, initial light intensity 500-800 μmol / m². 2 / s, the NaNO3 concentration in the culture medium is reduced to 0.08-0.12 g / L; the specific adjustment of DO regulation priority is as follows: when DO deviation occurs, first fine-tune the ventilation rate in an amplitude of ±0.02-0.04 vvm, and if it still cannot be corrected, then adjust it to 15-25 μmol / m 2 / s amplitude adjustment of light intensity; high frequency start RL training cycle is 8-16h, and astaxanthin characteristic fluorescence value is introduced as state input during training; the dissolved oxygen dynamic prediction basic model is corrected with tail gas oxygen content every set time, cycle is 2-4h, and tail gas oxygen content is collected by corrosion resistant infrared tail gas analyzer.

[0028] Optional, it also includes safety emergency procedures: setting up a three-level safety interlock mechanism:

[0029] If the DO is determined to be >150% or <30%, the first-level safety interlock is immediately triggered, cutting off the light, adjusting the ventilation rate to 1.0-1.2 vvm, and introducing sterile air containing 4%-6% CO2 until the DO returns to 50%-120%.

[0030] If the pipeline pressure decreases by more than 0.04-0.06 MPa within the set minimum time interval, or the conductivity of the culture medium increases by 4.5-5.5 mS / cm within the set minimum time interval, the second-level safety interlock will be activated to stop ventilation and lighting, locate the fault point, and replace the damaged parts.

[0031] If the cell density of *Rhodotorula purpureus* is determined to decrease continuously for 24 hours, the third-level safety interlock will be activated, and the model prediction controller will automatically switch to a low-energy mode with a light intensity of 250-350 μmol / m². 2The ventilation rate is 0.2-0.3 vvm / s, and 0.08-0.12 g / L glucose is added to the culture medium as an emergency carbon source.

[0032] Optionally, the data acquisition frequency in steps 4 and 5 is 20-40 seconds / time, and the data preprocessing includes moving average filtering and 3σ principle outlier removal;

[0033] During the control process, all data is stored on the local server and in the cloud at a preset frequency. After a fault occurs, a backtracking analysis report is automatically generated, containing the data trajectory, model parameters, and control actions for the time prior to the fault. The reward function for the RL agent during the biomass accumulation phase is:

[0034] ;

[0035] in It is the immediate reward value of the reinforcement learning agent. It is the biological target weight coefficient. It is the energy consumption penalty weighting coefficient. It is the cost factor for light energy. It is the ventilation energy consumption coefficient. It is the cell density growth rate. This is the average light intensity setting value. It is the average ventilation rate setpoint;

[0036] During the astaxanthin synthesis stage, the reward function of the RL agent is updated as follows:

[0037] ;

[0038] in It refers to the astaxanthin yield.

[0039] In summary, the present invention has at least one of the following beneficial technical effects:

[0040] This invention provides a method for coupled regulation of dissolved oxygen in the cultivation of *Rhodotorula pulmonata* using a corrosion-resistant photoreactor. By constructing a photoreactor made of corrosion-resistant material and a matching regulation system, the interference of material leaching on the culture medium is effectively avoided, ensuring the stability of the *Rhodotorula pulmonata* cultivation environment.

[0041] Based on the ternary fitting equation of oxygen transfer coefficient calibrated by dynamic method with light intensity, cell density and oxygen production rate, a basic model for dynamic prediction of dissolved oxygen was constructed. By combining model predictive control (MPC) and reinforcement learning (RL) agent, precise coupled regulation of dissolved oxygen was achieved.

[0042] During the biomass accumulation stage, the light intensity and ventilation parameters were adjusted in a coordinated manner to ensure the efficient growth of Haematococcus pluvialis cells and inhibit the excessive formation of biofilm. After entering the astaxanthin synthesis stage, the astaxanthin yield was prioritized and energy consumption was optimized by resetting the regulatory parameters and using the reinforcement learning reward function.

[0043] The three-level safety interlock mechanism and post-cultivation optimization steps further enhance the stability of the cultivation process and the iterative capability of the control strategy. Ultimately, this method significantly improves the precision of dissolved oxygen control, increases the biomass and astaxanthin yield of *Haematococcus pluvialis*, reduces energy consumption per unit product, and achieves intelligent and adaptive regulation of *Haematococcus pluvialis* cultivation, providing reliable technical support for the industrial production of *Haematococcus pluvialis*. Attached Figure Description

[0044] Figure 1 This is a schematic flowchart of the dissolved oxygen coupling regulation method for Haematococcus pluvialis culture in a corrosion-resistant photoreactor according to the present invention. Detailed Implementation

[0045] The present invention will be further described in detail below with reference to the accompanying drawings.

[0046] This invention discloses a method for coupled dissolved oxygen regulation in the cultivation of *Rhodotorula pulmonata* for use in corrosion-resistant photoreactors.

[0047] Reference Figure 1 Example 1, a method for coupled dissolved oxygen regulation in the cultivation of *Rhodotorula pulmonata* for corrosion-resistant photoreactors, includes the following steps:

[0048] Step 1: Construct a photoreactor and its supporting control system made of corrosion-resistant materials. The supporting control system includes a corrosion-resistant DO electrode, a gas mass flow controller, a quartz-encapsulated photosensor, and an ultrasonic biofilm monitoring sensor.

[0049] Step 2: Prepare a modified BG-11 medium for Haematococcus pluvialis and inject it into the photoreactor; calibrate the oxygen transfer coefficient under different aeration rates using a dynamic method, and establish a basic model for dynamic prediction of dissolved oxygen by fitting a ternary equation of light intensity, cell density, and oxygen production rate based on the photosynthetic characteristics of Haematococcus pluvialis.

[0050] Step 3: Introduce the logarithmic-stage Haematococcus pluvialis seed liquid into the photoreactor and simultaneously start the model prediction controller and reinforcement learning agent;

[0051] Step 4: When Haematococcus pluvialis is in the biomass accumulation stage, collect data on dissolved oxygen (DO), light intensity, aeration parameters, biofilm status, and culture medium characteristics. After data preprocessing, transmit the data to the control unit. Periodically use monitoring data to calibrate the dissolved oxygen prediction model, and adjust the light intensity and aeration parameters in a coordinated manner according to preset logic based on DO deviation. Periodically start RL training, update MPC control parameters based on culture performance, and perform biofilm cleaning operations as needed.

[0052] Step 5: When it is determined that Haematococcus pluvialis has reached the preset biomass threshold and entered the astaxanthin synthesis stage, reset the DO control interval, model parameters and reward function of the reinforcement learning agent of the model prediction controller, and simultaneously set the induction period culture conditions; correct the dissolved oxygen dynamic prediction basic model based on the metabolic characteristics of Haematococcus pluvialis during the induction period, and adjust the DO regulation priority; start RL training at high frequency to dynamically optimize the DO control interval and reward function weights.

[0053] By adopting the above technical solution, a photoreactor and a supporting control system made of corrosion-resistant materials are first built. By adapting the materials, corrosion and ion dissolution are avoided, ensuring a stable culture environment and accurate sensor measurements, thus providing a hardware foundation for dissolved oxygen control.

[0054] Subsequently, a modified culture medium specifically for Haematococcus pluvialis was prepared and injected into the reactor. The oxygen transfer coefficient under different aeration rates was determined by a dynamic method. A ternary correlation equation between light intensity, cell density, and oxygen production rate was constructed based on the photosynthetic characteristics of Haematococcus pluvialis, forming a basic model for dynamic prediction of dissolved oxygen that can quantify the changes in dissolved oxygen (DO), providing data support for regulation.

[0055] After the logarithmic-phase Haematococcus pluvialis seed liquid is introduced into the reactor, the Model Predictive Controller (MPC) and the Reinforcement Learning (RL) agent are simultaneously activated, enabling the control system to have real-time decision-making and autonomous learning capabilities.

[0056] When Haematococcus pluvialis is in the biomass accumulation stage, key data such as dissolved oxygen (DO) and light intensity are collected and preprocessed in real time. Based on the calibrated dissolved oxygen prediction model, light intensity and ventilation parameters are adjusted in a coordinated manner according to preset logic to stabilize DO. The RL agent is trained regularly to update the MPC parameters according to the culture effect. At the same time, the biofilm is cleaned as needed to ensure the precision of regulation and the culture environment.

[0057] When Haematococcus pluvialis reaches the preset biomass and enters the astaxanthin synthesis stage, the DO control range, model parameters, and RL reward function of MPC are reset to match the metabolic characteristics of this stage. After correcting the dissolved oxygen prediction model, the DO regulation priority is adjusted (with priority given to fine-tuning ventilation). The RL agent is trained at high frequency to dynamically optimize the regulation strategy, adapt to the high requirements of DO stability for astaxanthin synthesis, and achieve precise staged dissolved oxygen regulation.

[0058] Example 2 also includes a post-culture optimization step: when the astaxanthin content is determined to be greater than or equal to the cell dry weight threshold, the DO control accuracy, product yield and energy consumption index of this culture are evaluated after the culture is completed, and the culture data is added to the dissolved oxygen prediction model library and the reinforcement learning agent experience library; global RL training is carried out periodically to optimize the general parameters of cross-batch regulation and realize the self-iteration of the regulation strategy.

[0059] By adopting the above technical solution, the astaxanthin content reaches the cell dry weight threshold as the criterion for determining the end of culture, ensuring that the optimization process is started only after the culture target is achieved, and avoiding data interference with the optimization effect when the production target has not been reached.

[0060] After the cultivation process, the effects of DO control, product yield, and energy consumption were evaluated to quantify the regulatory effectiveness and resource consumption of the cultivation. Valuable regulatory data, such as effective DO control ranges and parameter combinations with low energy consumption and high output, were then selected. Adding this data to the dissolved oxygen prediction model library enriches the model's training samples and improves the model's prediction accuracy for DO changes under different cultivation scenarios. Adding it to the reinforcement learning (RL) agent experience library provides RL with more state-action-reward correlation samples, broadening its policy learning scope.

[0061] By periodically conducting global RL training, unlike the phased training during the cultivation process, global training integrates multiple batches of cultivation data, overcomes the limitations of single batch data, and optimizes cross-batch general control parameters in a targeted manner, such as the common biomass-astaxanthin stage DO switching threshold and the initial weights of the model predictive controller in different batches.

[0062] With continuous data accumulation and global training iteration, the control system can autonomously correct deviation parameters and optimize strategy logic, gradually forming a better control scheme that adapts to multiple batches of culture. Ultimately, it achieves self-iteration of the control strategy without human intervention, which promotes continuous improvement in DO control accuracy and product yield in subsequent batches of culture, while reducing energy consumption per unit product.

[0063] Example 3: The corrosion-resistant material mentioned in step 1 is quartz glass or polyvinylidene fluoride; the gas mass flow controller is made of 316L stainless steel; the outer shell of the corrosion-resistant DO electrode is made of titanium alloy and the sealing element is made of fluororubber; the thickness of the light-transmitting layer of the quartz-encased photosensor is greater than or equal to 2mm, ensuring compatibility with the weakly acidic or high-salt culture medium environment of Haematococcus pluvialis culture and eliminating the risk of material leaching.

[0064] By adopting the above technical solutions, and targeting the weakly acidic or high-salt culture medium environment for Haematococcus pluvialis cultivation, specific corrosion-resistant materials are selected for the photoreactor and its supporting control components, thereby avoiding the risk of material leaching and corrosion from the hardware source.

[0065] The photoreactor uses quartz glass or polyvinylidene fluoride, both of which have excellent tolerance to weak acids and high salts. Long-term contact with the culture medium will not cause ion leaching, thus avoiding interference with the metabolism of Haematococcus pluvialis. The gas mass flow controller uses 316L stainless steel, which has strong corrosion resistance and can prevent the inner wall from rusting due to the vapor of the culture medium or salt in the airflow, ensuring the accuracy of aeration control. The corrosion-resistant DO electrode uses titanium alloy as the shell and fluororubber as the seal. Titanium alloy is resistant to acids and alkalis and does not leach metal ions, while fluororubber can resist the corrosion of the culture medium, ensuring that the electrode can stably measure DO. The light-transmitting layer of the quartz-encased photosensor is no less than 2mm thick, which not only ensures the transmittance of photosynthetically active light for accurate monitoring of light intensity, but also enhances the material's corrosion resistance through sufficient thickness, preventing damage to the light-transmitting layer or the leaching of impurities.

[0066] The materials used for all components are matched to the characteristics of the Haematococcus pluvialis culture environment, achieving the effect of no material leaching and no component corrosion, providing a stable hardware foundation for subsequent precise control of dissolved oxygen and normal Haematococcus pluvialis culture.

[0067] Example 4, the composition of the modified BG-11 culture medium specifically for Haematococcus pluvialis in step 2 is as follows:

[0068] The culture medium is prepared with 1.2 g / L to 1.8 g / L NaNO3, 0.4 g / L to 0.6 g / L NaHCO3, and 0.06 g / L to 0.09 g / L MgSO4·7H2O. The pH of the culture medium is adjusted to 7.5 to 8.5. Sterilization is achieved by filtration using a polyethersulfone membrane, and a corrosion-resistant PE storage tank is used during the preparation process.

[0069] In Example 5, step 2, the dynamic method for calibrating the oxygen transfer coefficient specifically involves: introducing N2 into the photoreactor under algae-free conditions to reduce DO to below 10%, then switching to air, recording the DO recovery curve, and fitting the aeration rate kLa correlation equation:

[0070] kLa = 0.04 - 0.06 × Gas_set + 0.01 - 0.03; where Gas_set is the aeration rate setting; representing the volume of gas introduced per minute per unit volume of culture medium;

[0071] The ternary equation fitting the light intensity, cell density, and oxygen production rate is as follows:

[0072] ;

[0073] in This is the actual oxygen production rate of Haematococcus pluvialis. This is the maximum oxygen production rate of Haematococcus pluvialis. It is the initial light intensity of the photoreactor. is the light intensity half-saturation constant, k is the light absorption coefficient of *Rhodotorula purpureus* cells, C is the *Rhodotorula purpureus* cell density, and L is the optical path length of the photoreactor. It is the cell density half-saturation constant. It is the system correction coefficient. It is a nutrient limitation correction factor, determined by measuring the nitrogen concentration in the culture medium and substituting it into the empirical formula: ,in The concentration of inorganic nitrogen that can be directly utilized by Haematococcus pluvialis in the culture medium. These are characteristic parameters of nitrogen nutrition.

[0074] By adopting the above technical solution and determining the specific component concentration range of the modified BG-11 medium for Haematococcus pluvialis, suitable nitrogen, carbon, and magnesium nutrients are provided for the growth and photosynthesis of Haematococcus pluvialis: NaNO3 provides nitrogen source to support cell division and chlorophyll synthesis, NaHCO3 supplements carbon source to meet photosynthetic needs, and MgSO4·7H2O ensures enzyme activity and photosynthetic electron transfer.

[0075] The pH of the culture medium was adjusted to 7.5-8.5 to match the suitable acid-base environment for *Haemaphysalis*, avoiding excessive acidity or alkalinity that could inhibit its metabolism and photosynthetic efficiency. A polyethersulfone membrane filter was used for sterilization, removing microbial impurities at room temperature and preventing high-temperature sterilization from destroying heat-sensitive nutrients in the culture medium. Corrosion-resistant PE storage tanks were used during preparation to prevent ion leaching from the tank material from contaminating the culture medium, ensuring the purity, nutrient sufficiency, and environmental suitability of the culture medium, thus laying a high-quality nutritional foundation for subsequent *Haemaphysalis* cultivation.

[0076] The oxygen transfer coefficient was calibrated using a dynamic method: under algae-free conditions, N2 was first introduced to reduce DO to a low level, and then air was switched to allow DO to rise. By recording the rise curve, the oxygen transfer capacity of the reactor under different aeration rates was quantified, and the correlation equation between aeration rate and kLa was obtained by fitting, providing oxygen transfer characteristic data support for subsequent calculation of oxygen transfer rate and precise control of DO.

[0077] A ternary equation fitting the parameters of light intensity, cell density, and oxygen production rate was constructed. By introducing parameters such as the maximum oxygen production rate and the light intensity half-saturation constant, the synergistic effect of light intensity and cell density on photosynthetic oxygen production in *Rhodotorula purpureus* was accurately described. Systematic correction coefficients mitigated the interference of minor environmental fluctuations on oxygen production, and a nutrient limitation correction factor, calculated in conjunction with nitrogen concentration, reflected the regulatory role of nitrogen sources in oxygen production. The resulting ternary equation, together with the kLa correlation equation, forms the basis for dynamic prediction of dissolved oxygen based on quantifiable DO changes, providing core mathematical support for subsequent precise DO regulation.

[0078] Example 6, the preset logic in step 4 is: when DO is higher than the upper limit of the control range for the biomass accumulation stage, use 40-60 μmol / m 2 If the light intensity is reduced by an increment of 0.04-0.06 vvm, and the DO is still found to be above the limit, the ventilation rate is reduced by an increment of 0.04-0.06 vvm. If the DO is below the lower limit of the control range, the ventilation rate is increased by an increment of 0.04-0.06 vvm. If the DO is still not up to standard, the ventilation rate is increased by an increment of 40-60 μmol / m.2 / s amplitude increases light intensity.

[0079] Example 7: The biofilm cleaning operation is as follows: when the ultrasonic biofilm monitoring sensor detects a membrane thickness ≥50μm, first purge with 0.08-0.12MPa sterile compressed air for 10-15min. If the DO measurement deviation is still ≥10%, inject 0.08%-0.12% food-grade citric acid solution for circulation rinsing for 5-8min. Finally, replace with sterile culture medium.

[0080] By adopting the above technical solution, during the biomass accumulation stage of *Rhodotorula purpureus*, a priority-based coordinated adjustment strategy is used to stabilize dissolved oxygen (DO) when it deviates from the control range. When DO is above the upper limit, light intensity is reduced by a specific amount, as light intensity directly affects photosynthetic oxygen production by *Rhodotorula purpureus*, and this operation can quickly reduce the oxygen production rate. If DO still exceeds the limit, the aeration rate is further reduced to decrease external oxygen supply. When DO is below the lower limit, the aeration rate is increased to increase oxygen transfer, as aeration adjustment has a more direct impact on DO and energy consumption is controllable during the biomass accumulation stage. If DO still does not meet the standard, the light intensity is increased to enhance photosynthetic oxygen production. By clearly defining the adjustment priorities and magnitudes, efficient biomass accumulation is ensured while drastic fluctuations in DO are avoided, achieving precise dissolved oxygen control.

[0081] To address the interference of excessive biofilm formation (≥50 μm thickness) on light transmission and sensor measurements, a progressive cleaning strategy was adopted. First, sterile compressed air at a specific pressure was introduced for purging, physically removing the loosely attached biofilm through the impact of the airflow. This operation was gentle and did not disrupt the culture environment. If the DO measurement deviation remained significant after purging (DO ≥ 10%), a low-concentration food-grade citric acid solution was circulated for rinsing. This utilized the weakly acidic environment to dissolve the stubborn biofilm while avoiding damage to algal cells from high-concentration acid. Finally, the culture system was replaced with sterile culture medium to restore its purity. This progressive operation of physical purging, chemical cleaning, and system restoration effectively removed the biofilm, ensuring the measurement accuracy of the light sensor and DO electrode, and maintaining the reactor's efficient operation.

[0082] Example 8, the induction period culture conditions in step 5 include: temperature 26-30℃, initial light intensity 500-800 μmol / m². 2 / s, the NaNO3 concentration in the culture medium is reduced to 0.08-0.12 g / L; the specific adjustment of DO regulation priority is as follows: when DO deviation occurs, first fine-tune the ventilation rate in an amplitude of ±0.02-0.04 vvm, and if it still cannot be corrected, then adjust it to 15-25 μmol / m 2 / s amplitude adjustment of light intensity; high frequency start RL training cycle is 8-16h, and astaxanthin characteristic fluorescence value is introduced as state input during training; the dissolved oxygen dynamic prediction basic model is corrected with tail gas oxygen content every set time, cycle is 2-4h, and tail gas oxygen content is collected by corrosion resistant infrared tail gas analyzer.

[0083] By adopting the above technical solution, after Haematococcus pluvialis enters the astaxanthin synthesis induction period, specific culture conditions are set to meet the metabolic needs of this stage: the temperature is controlled at 26-30℃ to enhance the activity of astaxanthin synthase, and the initial light intensity is set at 500-800 μmol / m². 2 The system provides moderate light stress to trigger secondary metabolism, reduces the NaNO3 concentration in the culture medium to 0.08-0.12 g / L to create a nitrogen-limited environment, and the three factors work synergistically to induce efficient astaxanthin synthesis.

[0084] To address the sensitivity of astaxanthin synthesis during the induction period to light intensity fluctuations, the priority of DO regulation was adjusted: when DO deviates, the ventilation rate was first finely adjusted, as ventilation adjustment has less interference with photosynthetic metabolism and can avoid sudden changes in light intensity inhibiting the activity of synthases; if ventilation fine adjustment cannot correct DO, the light intensity was then adjusted even more significantly to minimize the impact of light environment fluctuations on product synthesis.

[0085] Because the metabolic state changes rapidly during the induction period, the RL training cycle is shortened to 8-16 hours to achieve high-frequency optimization. At the same time, the characteristic fluorescence value of astaxanthin is introduced as a state input, so that the RL agent can sense the product synthesis progress in real time and accurately adjust the regulation strategy to match the astaxanthin synthesis requirements.

[0086] Oxygen content in exhaust gas is collected every 2-4 hours using a corrosion-resistant infrared exhaust gas analyzer. Oxygen content in exhaust gas can directly reflect the actual photosynthetic oxygen production rate of Haematococcus pluvialis. This data is used to correct the basic model for dynamic dissolved oxygen prediction, ensuring that the model can match the changes in metabolic characteristics during the induction period in real time, improving the accuracy of DO prediction, and providing a guarantee for the accuracy of subsequent regulatory commands.

[0087] Example 9 also includes a safety emergency procedure: setting up a three-level safety interlock mechanism.

[0088] If the DO is determined to be >150% or <30%, the first-level safety interlock is immediately triggered, cutting off the light, adjusting the ventilation rate to 1.0-1.2 vvm, and introducing sterile air containing 4%-6% CO2 until the DO returns to 50%-120%.

[0089] If the pipeline pressure decreases by more than 0.04-0.06 MPa within the set minimum time interval, or the conductivity of the culture medium increases by 4.5-5.5 mS / cm within the set minimum time interval, the second-level safety interlock will be activated to stop ventilation and lighting, locate the fault point, and replace the damaged parts.

[0090] If the cell density of *Rhodotorula purpureus* is determined to decrease continuously for 24 hours, the third-level safety interlock will be activated, and the model prediction controller will automatically switch to a low-energy mode with a light intensity of 250-350 μmol / m². 2The ventilation rate is 0.2-0.3 vvm / s, and 0.08-0.12 g / L glucose is added to the culture medium as an emergency carbon source.

[0091] By adopting the above technical solutions, a three-level safety interlock mechanism is set up to achieve layered emergency protection against different types of risks in the cultivation of Haematococcus pluvialis, ensuring the stability of the cultivation system and cell viability.

[0092] When DO reaches extreme values, exceeding 150% or falling below 30%, the first level of safety interlock is triggered: cutting off light can quickly reduce oxygen production by Haematococcus pluvialis photosynthesis, preventing DO from continuing to rise; adjusting the ventilation rate to 1.0-1.2 vvm and introducing sterile air containing 4%-6% CO2 can both regulate DO concentration by enhancing airflow, accelerating oxygen escape when DO is high and increasing oxygen transfer when DO is low, and supplementing carbon sources to maintain cellular basal metabolism until DO returns to a safe range.

[0093] When pipeline pressure drops suddenly, posing a risk of leakage, or when the conductivity of the culture medium increases suddenly, posing a risk of material leaching, the second-level safety interlock is activated: stopping ventilation and light exposure can prevent the leakage from expanding or the leached ions from further contaminating the culture medium. By locating the fault point and replacing the damaged parts, the potential equipment and material hazards are eliminated from the source, preventing the risk from spreading and affecting the culture system.

[0094] When the density of Haematococcus pluvialis cells decreases continuously for 24 hours, indicating a risk of apoptosis, the third-level safety interlock is activated: the model prediction controller switches to a low-energy mode, reduces light intensity and ventilation rate to reduce the metabolic burden on cells; glucose is supplemented as an emergency carbon source to provide extra energy to support basic physiological activities, help cells restore activity, and avoid complete failure of the culture process.

[0095] The three-level interlocks correspond to dissolved oxygen physiological risks, equipment material risks, and cell survival risks, respectively. Through targeted emergency operations, risks can be quickly handled, ensuring the safety and stability of the entire Haematococcus pluvialis culture process.

[0096] In Example 10, the data acquisition frequency in steps 4 and 5 is 20-40 seconds / time, and the data preprocessing includes moving average filtering and 3σ principle outlier removal;

[0097] During the control process, all data is stored on the local server and in the cloud at a preset frequency. After a fault occurs, a backtracking analysis report is automatically generated, containing the data trajectory, model parameters, and control actions for the time prior to the fault. The reward function for the RL agent during the biomass accumulation phase is:

[0098] ;

[0099] in It is the immediate reward value of the reinforcement learning agent. It is the biological target weight coefficient. It is the energy consumption penalty weighting coefficient. It is the cost factor for light energy. It is the ventilation energy consumption coefficient. It is the cell density growth rate. This is the average light intensity setting value. It is the average ventilation rate setpoint;

[0100] During the astaxanthin synthesis stage, the reward function of the RL agent is updated as follows:

[0101] ;

[0102] in It refers to the astaxanthin yield.

[0103] By adopting the above technical solution, high-frequency data acquisition at 20-40s / times can capture the dynamic changes of key parameters such as DO and light intensity in a timely manner; moving average filtering is used to smooth data noise, and the 3σ principle is used to eliminate outliers, ensuring that the data input to the control unit is accurate and reliable, and providing a high-quality basis for control decisions.

[0104] During the control process, data is stored on the local server and the cloud at a preset frequency to achieve dual data backup to prevent loss. After a fault occurs, a retrospective report is automatically generated, which includes the data trajectory, model parameters and control actions set before the fault, making it easier to trace the cause of the fault and provide a reference for subsequent optimization.

[0105] The reward function of the reinforcement learning agent is designed in stages: In the biomass accumulation stage, the reward value is positively correlated with the cell density growth rate and negatively correlated with light energy and ventilation energy consumption. By weighting the biological objectives and energy consumption penalties, the agent is guided to prioritize optimizing rapid cell proliferation and controlling energy consumption. In the astaxanthin synthesis stage, the reward value becomes positively correlated with the astaxanthin yield. The weighting coefficient is adjusted to strengthen the incentive for product synthesis, guiding the agent to reduce resource consumption while ensuring efficient astaxanthin synthesis, thus achieving precise optimization of the core objectives at different culture stages.

[0106] The following specific embodiments illustrate the implementation principle of the present invention:

[0107] A 50L quartz glass photoreactor was constructed, with a supporting control system including: a corrosion-resistant DO electrode with a titanium alloy shell and fluororubber seal; a 316L stainless steel gas mass flow controller (MFC); a quartz-encased light sensor with a 2mm thick light-transmitting layer; and an ultrasonic biofilm monitoring sensor. The reactor and piping were first sterilized with saturated steam at 121℃ for 30 minutes, then rinsed with 0.5% peracetic acid circulation for 2 hours, and finally rinsed with sterile water until the pH reached 7.0 before use.

[0108] 2. Culture medium preparation and model establishment:

[0109] Prepare a modified BG-11 culture medium specifically for Haematococcus pluvialis: NaNO3 1.5 g / L, NaHCO3 0.5 g / L, MgSO4·7H2O 0.075 g / L. Adjust the pH to 8.0 with 1 mol / L NaOH. Sterilize by filtration through a 0.22 μm polyethersulfone filter membrane. After temporary storage in a corrosion-resistant PE storage tank, inject the medium into the reactor to a level of 40 L.

[0110] The oxygen transfer coefficient was calibrated using a dynamic method: Under algae-free conditions, N2 was introduced to reduce DO to 8%. After switching to air, the DO recovery curve was recorded, and the equation kLa = 0.05 × Gas_set + 0.02 was fitted (e.g., kLa = 0.035 when Gas_set = 0.3vvm). A ternary equation was fitted based on light intensity, cell density, and oxygen production rate.

[0111] ;

[0112] in =2.2 mg O2 / (Lh), =100μmol / (m 2 .s), =0.8gDW / L, =0.95; The initial nitrogen concentration of the culture medium [NO3] was determined. - Substituting [-N]=1.5mmol / L into f(N)=[N] / (0.2+[N]), we get f(N)=0.88.

[0113] Inoculation and System Startup:

[0114] Logarithmoidospora seed culture (density 1×10⁻⁶) 6 (cells / mL) were inoculated into the reactor at a 10% inoculum rate, and the acclimation conditions were a temperature of 25℃ and a light intensity of 300 μmol / (m²). 2 (s), ventilation rate 0.3 vvm, cell specific growth rate μ = 0.22d after 48h. -1 The Model Predictive Controller (MPC) and PPO algorithm reinforcement learning (RL) agent were launched. The initial parameters of the MPC were: prediction step size 12 (5 min per step), control step size 3 (output instructions every 15 min), and DO control interval [70%, 90%]. The RL experience base was pre-filled with 3 batches of historical training data, and the initial weights of the reward function were α=8, β=2, γ=0.5, and λ=0.8.

[0115] Regulation of biomass accumulation phase:

[0116] Biomass accumulation stage (cell density 1×10⁻⁶) 6 -5×10 6 (cells / mL):

[0117] Data acquisition: DO, light intensity, ventilation rate, biofilm thickness, and conductivity are collected every 30 seconds. After 10-second moving average filtering and 3σ outlier removal, the data is transmitted to the control unit and stored locally and in the cloud every 1 minute.

[0118] DO regulation: When DO > 90%, first reduce the light intensity by 50 μmol / (m²). 2 .s) (minimum 200 μmol / (m 2 If the DO (dosage) still exceeds the standard, reduce the ventilation rate by 0.05 vvm (minimum 0.2 vvm); if DO < 70%, first increase the ventilation rate by 0.05 vvm (maximum 0.8 vvm), and if it still does not meet the standard, increase the light intensity by 50 μmol / (m²). 2 .s) (up to 1000 μmol / (m 2 .s)).

[0119] RL training and biofilm cleaning: RL training is started every 24 hours, and MPC weights are updated based on the cell density growth rate. When the ultrasonic sensor detects that the biofilm thickness is ≥50μm, it is purged with 0.1MPa sterile air for 12min. If the DO measurement deviation is still ≥10%, 0.1% food-grade citric acid solution is injected for circulation rinsing for 6min, and then 5L of sterile culture medium is used to replace it.

[0120] Regulation of astaxanthin synthesis:

[0121] When the cell density reaches 5×10 6 cells / mL (dry weight 3.8 g DW / L), entering the astaxanthin synthesis stage:

[0122] Induction conditions: temperature 28℃, initial light intensity 650 μmol / (m²) 2 .s), the NaNO3 concentration in the culture medium was reduced to 0.1 g / L ([NO3 - -N]=0.14mmol / L, f(N)=0.41).

[0123] MPC and RL Reset: The MPCDO control interval is narrowed to [80%, 90%], the prediction step size is 15, and the RL reward function is updated to R=10×dAsta / dt-3×(0.6×I_set_avg+0.9×Gas_set_avg). I_set_avg is the average light intensity setpoint, which refers to the arithmetic mean of the light intensity setpoints issued by the control system within a specific time period (such as the control period of the model predictive controller or the reward function calculation window). Gas_set_avg is the average ventilation rate setpoint, which refers to the arithmetic mean of the ventilation rate setpoints issued by the control system within the same time period. RL training is started every 12 hours, and the 470nm astaxanthin characteristic fluorescence value is introduced as the state input.

[0124] DO control and model correction: When DO deviation occurs, first fine-tune the ventilation rate by ±0.03 vvm (maximum 0.9 vvm). If this still fails to correct the deviation, adjust by ±20 μmol / (m²). 2 .s) Adjust the light intensity; collect the oxygen content of the exhaust gas every 3 hours using a corrosion-resistant infrared exhaust gas analyzer, and correct the dissolved oxygen dynamic prediction model (e.g., if the oxygen content of the exhaust gas drops from 21% to 19%, correct the OPR to 0.85 times the initial value).

[0125] Safety emergency response and post-training optimization:

[0126] Safety interlock: When DO rises to 160%, light is cut off, ventilation rate is adjusted to 1.1 vvm, and sterile air containing 5% CO2 is introduced. After 15 minutes, DO drops to 110%. When the pipeline pressure drops by 0.05 MPa within 10 minutes, ventilation and light are stopped. Inspection revealed a leak at the MFC pipeline interface. The gasket was replaced and the system was restarted. Cell density continuously decreased from 5 × 10⁻⁶ cells / day over 24 hours. 6 The number of cells / mL decreased to 4.2 × 10⁻⁶. 6 When the light intensity is 300 μmol / (m³ / mL), the MPC switches to low-energy mode (light intensity 300 μmol / (m³ / mL)). 2 With a ventilation rate of 0.25 vvm and supplementation with 0.1 g / L glucose, the cell density recovered to 4.8 × 10⁻⁶ cells / day after 48 hours. 6 cells / mL.

[0127] Post-culturing optimization: After 12 days of culture, the astaxanthin content reached 1.8% of the cell dry weight, and the culture was terminated. The DO control precision was evaluated as 92%, biomass yield as 4.2 g DW / L, astaxanthin yield as 75.6 mg / L, and energy consumption per unit of astaxanthin as 22 kWh / g. Data were added to the model library and RL experience library, and global RL training was conducted every 8 batches to optimize the DO switching threshold for the biomass-astaxanthin stage to 4.8 × 10⁻⁶. 6 cells / mL, MPC initial Q weight is 12.

[0128] Comparison of effects with traditional control measures

[0129] 1. Comparison benchmark (traditional scheme):

[0130] Reactor and components: 20L ordinary borosilicate glass reactor, ordinary glass shell DO electrode, carbon steel MFC, biofilm-free monitoring sensor.

[0131] Culture medium and control: ordinary BG-11 culture medium (sterilized at 121℃ for 20 min), single PID control DO (fixed setpoint 80%, Kp=5, Ti=300s, Td=50s), without staged control logic.

[0132] Auxiliary measures: No active cleaning of biofilm (manual weekly wiping of reactor inner wall), no system safety interlock (relying on manual inspection), no post-culture data iteration (fixed parameters are used).

[0133] The comparison results of key performance indicators are shown in Table 1:

[0134] Table 1

[0135]

[0136] Compared to traditional control methods, this solution achieves comprehensive performance improvements through a collaborative design that integrates corrosion-resistant hardware adaptation, phased intelligent control, and safe iterative optimization: In terms of control precision, DO control precision is increased from 70% to 92%, accurately matching the metabolic needs of Haematococcus pluvialis at different stages; in terms of product efficiency, biomass yield is increased by 40% and astaxanthin content is increased by 80%, breaking through the product accumulation bottleneck of traditional cultivation; in terms of energy consumption cost, energy consumption per unit of astaxanthin is reduced by 42%, significantly reducing resource waste; and in terms of stability and safety, the cultivation failure rate is reduced from 15% to 2%, and the batch-to-batch key indicator deviation is narrowed to ±5%, completely solving the problems of reliance on manual labor and large fluctuations in traditional solutions, providing stable and reliable technical support for the large-scale and efficient cultivation of Haematococcus pluvialis.

[0137] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for coupling regulation of Haematococcus cultivation dissolved oxygen in corrosion-resistant photoreactor, characterized in that, The method comprises the following steps: Step 1, build a photoreactor made of corrosion-resistant material and a matching control system; Step 2, prepare a special improved BG-11 medium for Haematococcus pluvialis and inject it into the photoreactor; Calibrate the oxygen transfer coefficient under different aeration rates by dynamic method, and combine the photosynthetic characteristics of Haematococcus pluvialis to fit a ternary equation of light intensity, cell density and oxygen production rate, to establish a dynamic prediction basic model of dissolved oxygen; Step 3, introduce the logarithmic phase seed liquid of Haematococcus pluvialis into the photoreactor, and simultaneously start the model prediction controller and the reinforcement learning intelligent agent; Step 4, when the Haematococcus pluvialis is in the biomass accumulation stage, collect the data of DO, light intensity, aeration parameters, biological membrane state and medium characteristics, and after data preprocessing, input them into the control unit; regularly correct the dissolved oxygen prediction model with the monitoring data, and according to the DO deviation, adjust the light intensity and aeration parameters according to the preset logic; periodically start RL training, update the MPC control parameters based on the culture performance, and perform biological membrane cleaning operation as needed; Step 5, when it is judged that the Haematococcus pluvialis reaches the preset biomass threshold and enters the astaxanthin synthesis stage, reset the DO control interval, model parameters of the model prediction controller and the reward function of the reinforcement learning intelligent agent, and simultaneously set the induction period culture conditions; correct the dissolved oxygen dynamic prediction basic model based on the metabolic characteristics of Haematococcus pluvialis in the induction period, and adjust the DO control priority; start RL training, and dynamically optimize the DO control interval and the weight of the reward function; In step 1, the matching control system includes a corrosion-resistant DO electrode, a gas mass flow controller, a quartz-wrapped light sensor and an ultrasonic biological membrane monitoring sensor; The composition of the medium in step 2 is: NaNO31.2g / L-1.8g / L, NaHCO30.4g / L-0.6g / L, MgSO4.7H2O 0.06g / L-0.09g / L, and the pH of the medium is adjusted to 7.5-8.5; In step 2, the dynamic calibration of the oxygen transfer coefficient is as follows: under the condition of no algae, N2 is introduced into the photoreactor to reduce the DO to below 10%, then air is switched, and the DO rising curve is recorded, and the aeration rate kLa correlation equation is fitted: kLa=0.04-0.06×Gas_set+0.01-0.03; wherein Gas_set is the set value of the aeration rate; The ternary equation is: ; wherein is the actual oxygen production rate of Rhodella reticulata, is the maximum oxygen production rate of Rhodella reticulata, is the initial light intensity of the photobioreactor, is the light intensity half-saturation intensity, k is the light absorption coefficient of Rhodella reticulata cells, C is the cell density of Rhodella reticulata, and L is the light path length of the photobioreactor, is the cell density half-saturation constant is the system correction coefficient, is the nutrient limitation correction factor, the nitrogen concentration in the culture medium is measured, and the empirical formula is substituted: wherein represents the concentration of inorganic nitrogen that can be directly utilized by Rhodella reticulata in the culture medium, is a characteristic parameter of nitrogen nutrition; The data acquisition frequency in steps 4 and 5 is 20-40s / time, and the data preprocessing includes sliding average filtering and 3σ principle outlier rejection; The reward function of the RL intelligent agent in the biomass accumulation stage is: ; wherein is an immediate reward value of the reinforcement learning agent, is a biological objective weight coefficient, is an energy consumption penalty weight coefficient, is a light energy cost coefficient, is a ventilation energy consumption coefficient, is a cell density growth rate, is an average light intensity set value, is an average ventilation rate set value; The reward function of the RL intelligent agent in the astaxanthin synthesis stage is updated as: ; wherein is astaxanthin yield.

2. The method for red ball algae cultivation dissolved oxygen coupling regulation and control for corrosion-resistant photo-reactor according to claim 1, characterized in that, It also includes an optimization step after culture: when it is judged that the astaxanthin content is greater than or equal to the cell dry weight threshold, the DO control accuracy, product yield and energy consumption indicators of this culture are evaluated after the culture is completed, and the data of this culture are supplemented to the dissolved oxygen prediction model library and the reinforcement learning intelligent agent experience library; periodically carry out global RL training to optimize the general parameters of cross-batch control, and realize the self-iteration of the control strategy.

3. The method for red ball algae cultivation dissolved oxygen coupling regulation and control for corrosion-resistant photo-reactor according to claim 2, characterized in that, The preset logic in Step 4 is: when DO is higher than the upper limit of the control interval in the biomass accumulation stage, reduce the light intensity by 40-60 μmol / m² / s, and if it is still judged that DO is over-standard, reduce the aeration rate by 0.04-0.06 VVM; when DO is lower than the lower limit of the control interval, increase the aeration rate by 0.04-0.06 VVM, and if DO still does not meet the standard, increase the light intensity by 40-60 μmol / m² / s.

4. The method for red ball algae cultivation dissolved oxygen coupling regulation and control for corrosion-resistant photo-reactor according to claim 3, characterized in that, The biofilm cleaning operation is when the ultrasonic biofilm monitoring sensor detects that the film thickness is ≥50 μm, first blow in 0.08-0.12 MPa sterile compressed air for 10-15 min, if it is judged that the DO measurement deviation is still ≥10%, then inject 0.08%-0.12% food-grade citric acid solution for 5-8 min, and finally replace it with sterile culture medium.

5. The method for red ball algae cultivation dissolved oxygen coupling regulation and control for corrosion-resistant photo-reactor according to claim 4, characterized in that, The induction period culture conditions in Step 5 include: temperature 26-30℃, initial light intensity 500-800 μmol / m² / s, and NaNO3 concentration in the culture medium reduced to 0.08-0.12 g / L; the DO control priority adjustment is: when the DO deviation, first fine-tune the aeration rate by ±0.02-0.04 VVM, and if it still cannot be corrected, then adjust the light intensity by 15-25 μmol / m² / s; the cycle of high-frequency start RL training is 8-16 h, and the astate input is introduced simultaneously during training; the cycle of correcting the dynamic prediction base model of dissolved oxygen with tail gas oxygen content every set time is 2-4 h, and the tail gas oxygen content is collected by a corrosion-resistant infrared tail gas analyzer.

6. The method for red ball algae cultivation dissolved oxygen coupling regulation and control for corrosion-resistant photo-reactor according to claim 5, characterized in that, It also includes a safety emergency step: a three-level safety interlock mechanism is set: If it is judged that DO > 150% or < 30%, the first level safety interlock is triggered immediately, the light is turned off, the aeration rate is adjusted to 1.0-1.2 VVM, and sterile air containing 4%-6% CO2 is introduced until DO returns to 50%-120%; If it is judged that the pipeline pressure decreases by more than 0.04-0.06 MPa within a set minimum time interval or the culture medium conductivity increases by 4.5-5.5 mS / cm within a set minimum time interval, the second level safety interlock is started, the aeration and light are stopped, the fault point is located and the damaged parts are replaced; If it is judged that the Haematococcus cell density has decreased continuously for 24 h, the third level safety interlock is started, the model predictive controller automatically switches to a low energy consumption mode, the light intensity is 250-350 μmol / m² / s, the aeration rate is 0.2-0.3 VVM, and 0.08-0.12 g / L of glucose is supplemented as an emergency carbon source to the culture medium.

Images