Photodynamic inactivation parameter adjusting method based on hybrid fuzzy-PID (Proportion Integration Differentiation) control

Abstract

The invention discloses a photodynamic inactivation parameter adjusting method based on hybrid fuzzy-PID control, relates to the technical field of parameter fuzzy control, and aims at the problem that photodynamic inactivation efficiency and aquatic organism safety in aquaculture are difficult to consider at the same time through multi-sensor fusion and environment prediction. Then, a fuzzy-PID self-adaptive putting technology is adopted to be coupled with an illumination strategy, and accurate matching of the photosensitizer dosage and the light intensity is achieved; water body fluctuation and pathogen diffusion are dealt with through real-time closed-loop and multi-scene collaborative scheduling; multi-dimensional index evaluation is utilized, reinforcement learning is introduced to carry out self-learning optimization, and putting and illumination parameters are continuously improved; and finally, the method is expanded to large-scale distributed control and cloud big data training, and efficient inactivation and flexible deployment can be kept in different breeding scenes. Pathogens can be effectively prevented from being rapidly diffused in an adverse environment, damage to aquatic animals caused by excessive accumulation of reactive oxygen species (ROS) is reduced, and meanwhile energy consumption and resource waste are reduced.

Description

Technical Field

[0001] The invention relates to the technical field of parameter fuzzy control, in particular to a photodynamic inactivation parameter adjustment method based on mixed fuzzy-PID control. Background Art

[0002] At present, water disinfection plays a vital role in municipal water supply, sewage treatment, industrial wastewater reuse and large-scale aquaculture systems. With the acceleration of urbanization and the continuous development of industrial production, the content of suspended solids, organic matter and pathogenic microorganisms in water bodies shows highly dynamic and nonlinear changes. Affected by multiple factors such as seasonal climate, rainfall, industrial emissions, aquaculture feeding and biological metabolism, water quality parameters (such as pH, dissolved oxygen, temperature and light intensity) often fluctuate violently. At the same time, modern water treatment systems are tending to adopt advanced sensor networks, distributed control and cloud data analysis technologies to monitor and dynamically control water status in real time to achieve more efficient and accurate disinfection and purification. Various types of intelligent control devices are constantly emerging, providing new ideas and new means for handling complex water quality and ensuring the safety of aquatic organisms, making water disinfection not only limited to traditional chemical treatment, but gradually shifting to environmentally friendly and energy-saving technologies such as photodynamic inactivation and photocatalysis.

[0003] A Chinese invention patent with application publication number CN101301538A discloses a fuzzy intelligent control system for photosensitizer dosing in water treatment. The method adopted is: first, based on the current state and operation amount of the controlled object, an online prediction model of the controlled object is established, and the control result of maintaining the current operation amount is predicted by this model. In the control target evaluation, the control effect of the system is evaluated based on the current state and prediction result of the controlled object, and the result of adjusting the fuzzy control weight is generated. The fuzzy control can determine the optimal operation increment accordingly.

[0004] However, in practical applications, existing water disinfection control systems generally face the problems of delayed response and inaccurate regulation. Especially when using photodynamic inactivation technology, the photosensitizer and light source irradiation parameters in the water must be strictly matched to produce enough reactive oxygen species (ROS) to achieve pathogen inactivation. However, traditional control strategies often use fixed set values ​​or simple PID control, which cannot respond in real time to sudden changes in key parameters such as pH, dissolved oxygen and temperature caused by changes in the natural environment, fluctuations in industrial loads or adjustments in aquaculture operations. In this case, too low photosensitizer dosage or insufficient light will lead to insufficient pathogen inactivation, while too high dosage or light may cause excessive ROS, thereby damaging aquatic animal tissues and destroying the ecological balance of the water body. More importantly, the inconsistent dimensions and nonlinear characteristics between multi-source sensor data make it difficult for traditional closed-loop control methods to simultaneously take into account inactivation efficiency, energy consumption and safety, which may ultimately lead to pathogen spread, unstable disinfection effect and a sharp increase in energy consumption.

[0005] To this end, the present invention provides a photodynamic inactivation parameter adjustment method based on hybrid fuzzy-PID control. Summary of the invention

[0006] 1. Technical issues to be resolved In view of the deficiencies of the prior art, the present invention provides a photodynamic inactivation parameter adjustment method based on hybrid fuzzy-PID control, which grasps the dynamics of water quality through multi-sensor fusion and environmental prediction; then uses fuzzy-PID adaptive delivery technology coupled with illumination strategy to achieve accurate matching of photosensitizer dosage and light intensity; then responds to water body fluctuations and pathogen spread through real-time closed-loop and multi-scenario collaborative scheduling; uses multi-dimensional indicator evaluation and introduces reinforcement learning for self-learning optimization, and continuously improves delivery and illumination parameters; finally, it is extended to large-scale distributed control and cloud-based big data training, which can maintain efficient inactivation and flexible deployment in different breeding scenarios. It can effectively prevent the rapid spread of pathogens in adverse environments and reduce the damage to aquatic animals caused by excessive accumulation of ROS, thereby solving the technical problems raised in the background technology.

[0007] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: The photodynamic inactivation parameter adjustment method based on hybrid fuzzy-PID control includes: when the pH value and dissolved oxygen O 2 When key indicators such as temperature and water quality fluctuate significantly at the same time, the system receives the output of multi-source sensors and performs error correction and redundancy detection on the information of each channel through weighted filtering to generate an environmental fusion data vector. Based on this vector, an adaptive prediction model is built to identify extreme water quality changes and output future short-term environmental trends for subsequent deployment and lighting strategy formulation, thereby issuing early warnings at the stage of water quality mutation; Complete the generation and output of environmental fusion data vectors, compare the optimal activity range of photosensitizers and the distribution of pathogens, use adaptive formula delivery technology to calculate the amount of photosensitizers delivered, and formulate a lighting plan based on the environmental matching coefficient and water body dynamic prediction results to strengthen irradiation at local hot spots and reduce energy consumption in normal areas. Through the coupling of delivery amount, delivery time, and irradiation, maximize the inactivation efficiency and reduce surface damage to aquatic animals, and avoid the accumulation of reactive oxygen species (ROS) caused by excessive light or excessive photosensitizers. After the photosensitizer has been released and illumination has begun, if the sensor detects that the environmental condition data deviates significantly from the target again, the mixed fuzzy-PID real-time closed-loop adjustment is performed, and the scene difference vector is used to adapt to the water body regulation needs. When the water environment fluctuates violently or the risk of pathogen spread increases, the local irradiation is strengthened or the isolation device is activated in time. After each control cycle, the deviation vector and compensation instruction are retained as data; When the photodynamic inactivation process ends or reaches the expected duration, the treatment effect is summarized and evaluated based on the aforementioned compensation instructions and control feedback indicators, and the replay pool mechanism in deep reinforcement learning is used to train each state-action-reward sequence, update the adaptive delivery and lighting strategy model, and combine the data curve generated in the evaluation process with the scene difference vector to summarize the key environmental influencing factors and solidify the optimized parameters into the knowledge base; In large-scale farming scenarios with many partitions or across regions, local closed-loop scheduling is achieved by deploying edge control units in each sub-area, and environmental fusion data and control command logs are regularly uploaded to the cloud. Multi-objective training is performed using deep learning algorithms to form a common decision-making model, which is then sent to each edge control unit for synchronous execution.

[0008] Preferably, the sensor array deployed in the aquaculture water body collects environmental condition parameters at a high frequency, performs redundancy detection and error correction on multi-source sensor data, and then uses a custom fusion formula to form a fusion output The fusion formula is based on the exponential decay mechanism of the difference between sensors. The above fusion processing is performed on each of the environmental condition parameters respectively to obtain the fused data Rh.

[0009] Preferably, after obtaining the output fusion data, an adaptive prediction model is used to construct a dynamic environment model , using deep neural network or fuzzy neural network to predict the trend of environmental condition parameters in the short term; multi-channel input of historical data of each environmental condition parameter and its fusion value; At time t, the fused data Rh and the fused values ​​of several past time points are input into the trained or deployed neural network to obtain the multi-parameter environmental prediction information for the next period; After each prediction is completed, the prediction results are stored in the environmental prediction database and the environmental state model is generated , if it is detected that certain parameters are about to exceed the safety threshold, early intervention will be carried out; Preferably, based on the environmental prediction information obtained, the precise amount and timing of the photosensitizer delivery are determined, and the basic delivery amount of the photosensitizer is defined. and environmental sensitivity parameters; when it is predicted that the environmental condition parameters will significantly deviate from the ideal range of the photosensitizer, the dosage will be appropriately reduced or the delivery will be delayed; Introducing environmental coordination factor , used to represent the future The suitability of the water body for the administration of photosensitizer after a certain period of time; In order to obtain the environmental matching coefficient Then, calculate the amount of photosensitizer , and implement the corresponding delivery according to the working mode of the delivery equipment. If it is too low, delay the time of release appropriately, or call in auxiliary equipment to correct the water quality before release.

[0010] Preferably, after completing the amount of photosensitizer added After the determination and execution of and other auxiliary indicators, and then formulate a lighting strategy; the lighting strategy needs to ensure the matching degree between the photosensitizer and the light, and carry out refined irradiation of the identified local hot spots, and strengthen the irradiation of areas with high pathogen concentrations through mobile light sources or optical fiber arrays; Define lighting optimization function , used to determine the irradiation intensity The duration of irradiation After the function is executed, the actual distribution state of the photosensitizer is used as the input variable, and the environmental matching coefficient is combined with the optimal matching relationship between the two. , which can be described as:

[0011] Where: is the base coefficient of the lighting strategy, is the amount of photosensitizer added; , are sensitivity amplification factors respectively; Light intensity and duration of light exposure Functions can be optimized based on lighting Obtained under the following scheduling principles: When the lighting optimization function When the light intensity is higher than expected, a higher light intensity is automatically selected. Prolong the duration of light exposure ; When the lighting optimization function When it is not higher than the expected value, it tends to reduce the light intensity And appropriately shorten the duration of light exposure ; Lighting scheduling can be solved by offline or online optimization algorithms, and can also be fine-tuned by a microprocessor in the subsequent real-time closed-loop stage.

[0012] When implementing actual irradiation, spatial zoning will be carried out according to the distribution of pathogens: The hotspot area is illuminated with relatively high intensity, and the non-hotspot area is illuminated with normal intensity or no illumination; the time interval between the release and illumination is determined by the illumination optimization function. The photosensitizer concentration reflected in the formula is determined by the degree of coordination; when the expected photosensitizer is fully diffused in the water body and has not significantly attenuated, the illumination can be started; Preferably, based on the results of photosensitizer placement and illumination parameter setting, and comparing the real-time sensor data of the actual environment, if a significant deviation between the current state and the set target interval is observed, a closed-loop correction mode is entered; Get new environmental measurements at time t to form an instant state vector , and the ideal interval vector Compare and get the environmental deviation vector :

[0013] A control method based on nonlinear potential function is used to transform the environmental deviation vector Mapping to control signal vector , including compensation for light intensity and start and stop scheduling of auxiliary devices , which can be written as:

[0014] in, is a nonlinear potential function; It is a mapping operation, which is used to convert the output of the potential function into a command value that can directly drive the actuator; Light compensation: If light intensity compensation If it is positive, the light source power is increased; if it is negative, the light source power is reduced or the effective irradiation time is shortened ; Auxiliary device scheduling: auxiliary device start and stop scheduling It can trigger oxygenation equipment, pH buffer dosing unit, local water flow isolation device, etc., and adaptively control according to the source of deviation; Preferably, according to the previously determined illumination mode and photosensitizer distribution plan, combined with a new round of sensor data and closed-loop control instructions, coordinated scheduling is performed in spatial and temporal dimensions to ensure that the ideal inactivation effect and aquatic animal safety can be achieved in each scenario; Define the environment difference vector , whose components may include water flow velocity, stocking density, and light interference level; Build multi-scenario scheduling functions , the control signal vector Difference vector with environment Perform joint mapping to obtain scenario-based instructions : When the water flow rate When it is large, multiple scene scheduling functions Compensation for light intensity Do partition enlargement to compensate for the decrease in photosensitizer concentration caused by water dilution; When the stocking density When it is large, multiple scene scheduling functions Auxiliary device start and stop scheduling can be called for auxiliary devices Make priority enhancements; If the light interference Too high, multi-scenario scheduling function Automatically suppress excessive increase in light intensity and extend lighting time ; Scenario-based instructions Finally, the scheduling module sends it to each independent execution subsystem for execution: Preferably, a systematic multi-dimensional evaluation is conducted based on the sensor records, control instruction records and final water pathogen inactivation results generated during the actual implementation process, against the set target indicators, mainly including the following contents: Record pathogen clearance rate through fluorescent imaging device or other pathogen detection module , by observing and detecting surface damage or other health indicators of aquatic animals, and recording the injury rate of individuals ; Record data on oxygen enrichment equipment, isolation devices, light source power, etc. to obtain comprehensive energy consumption indicators ; and monitor after treatment to The recovery of pH, dissolved oxygen, temperature, etc. within a certain period of time, and the effect evaluation report is formed and output based on the above three types of evaluation results; Preferably, the output effect evaluation report is combined with the recorded execution actions to construct an adaptive optimization mechanism, which is based on a state-action-reward logic cycle to provide control and improvement for each round of photodynamic inactivation process, wherein: Defining states and define actions ; To balance pathogen inactivation efficiency with aquatic animal safety and energy consumption, a comprehensive reward function is defined ; The larger the value, the better the balance between inactivation efficiency, aquatic animal safety and energy consumption. Otherwise, the corresponding strategy should be revised in the next round of training. Stored in the replay pool, the mapping strategy is formed after iterative training ; Preferably, the policy model parameters and key evaluation data obtained from this round of training are stored in the knowledge base: the output of the reinforcement learning model is stored separately according to different scenario labels, and the mapping relationship of each key variable is retained to facilitate rapid matching of similar scenarios; in certain extreme situations, expert rules and reinforcement learning models are allowed to be used in parallel and stored in the same knowledge base. If an extreme state occurs, the expert rules are triggered first to prevent serious accidents; Before a new batch of breeding begins, the corresponding strategy parameter set is automatically loaded according to the scene label, and the initial photosensitizer dosage is preset. , Environmental coordination coefficient Thresholds and lighting optimization plans, if there are major environmental changes, additional fine-tuning will be initiated; Preferably, an edge control unit is deployed in each sub-area to quickly perform the following operations in a local area: Each edge control unit receives real-time environmental condition data within the partition and uses an updated expert knowledge base or reinforcement learning strategy to adjust the amount of photosensitizer delivered. , irradiation intensity The duration of irradiation Equivalent feedback control strategies for fast adaptive adjustment at the edge level; If the environmental coordination factor Or if there is a local extreme deviation in the distribution of pathogens within a sub-division, the edge control unit will respond quickly under low latency conditions; when multiple sub-divisions are adjacent, the edge control units will communicate key status information with each other and coordinate the implementation of cross-division isolation or water flow regulation within the sub-division; Each edge control unit makes the first-level quick decision in the local environment, and transmits the phased summarized data to the main control center, which integrates the multi-partition information.

[0015] Preferably, the environmental parameters, control actions and evaluation indicators generated by each edge control unit when running locally are regularly uploaded to the cloud big data platform and supplemented with auxiliary information; Build a multi-objective optimization deep learning framework on the cloud platform, including a multi-task network structure or a multi-head output layer; Perform large-scale batch training on the collected data based on the generalization function: After training is completed, the cloud will use the new cloud decision model Send it to each edge control unit; If the farming environment of a sub-district or region is highly matched with the cloud training data, the cloud decision model can be directly adopted If there are still local differences, secondary fine-tuning can be performed locally in the edge control unit based on the reinforcement learning mechanism; Optimally, based on cloud-based decision-making models Combined with the data from the partitioned edge control units, cross-scenario data fusion and global strategy iteration are performed for different farming scenarios, and the global optimal strategy is regularly formed and fed back to the self-learning process; In the proposed environmental difference vector The environmental difference vector reported by different edge control units is analyzed by quantifying factors such as flow rate, breeding density, light interference, etc. Perform unified processing and compare with the cloud model to identify possible blind spots of scene differences, add new feature dimensions or modify existing feature weights; The above scenario difference information is combined with the generated cloud decision model Combination, using distributed strategy iteration algorithm:

[0016] Where: It is a distributed fusion function used to aggregate the feedback results of each edge control unit and then perform the cloud decision model Fine-tune and generate a global decision model .

[0017] (III) Beneficial effects The present invention provides a photodynamic inactivation parameter adjustment method based on hybrid fuzzy-PID control, which has the following beneficial effects: The photosensitizer dosage and light parameters are highly coupled and dynamically optimized. Through the hybrid fuzzy-PID algorithm, nonlinear adjustments can be made quickly in response to fluctuations in water environmental conditions, which not only avoids the waste of resources caused by traditional fixed ratios, but also reduces the damage and pollution to aquatic animals in extreme environments. Environmental coordination coefficients and environmental difference vectors are fully utilized in real-time closed-loop control and multi-scenario scheduling: they seamlessly connect the obtained prediction models and release strategy parameters with online decision-making, and can achieve differentiated disinfection and water pollution treatment through local water flow isolation or enhanced irradiation in hot spots in scenarios such as indoor high density or cage mobility, thereby enhancing system adaptability; The reinforcement learning mechanism is introduced into the effect evaluation and self-learning optimization to balance the inactivation rate of pathogens, the damage rate of aquatic animals and energy consumption. Through the comprehensive evaluation of playback pool training and multi-objective reward function, the continuous iterative improvement of the feedback control strategy is achieved. The linkage between the edge control unit and the cloud platform gives the solution strong scalability: each edge control unit can execute the knowledge base and cloud decision model locally, and the edge quickly closes the loop; the cloud integrates the current state data and control feedback data for deep training, extracts the general strategy or model and then sends it again to complete the global strategy iteration. This layered collaborative approach makes hybrid fuzzy-PID and reinforcement learning no longer limited to a single water body, but can cover diversified applications across regions or seasons. To sum up the above, the combination of technical elements and synergistic efficiency have been achieved in the aspects of photosensitizer delivery and light control, environmental perception and real-time scheduling, as well as self-learning and distributed collaboration, which has greatly improved the accuracy, response speed and ecological safety of photodynamic inactivation of aquatic products. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The present invention is a schematic flow chart of a method for adjusting photodynamic inactivation parameters based on hybrid fuzzy-PID control. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1 The present invention provides a method for adjusting photodynamic inactivation parameters based on hybrid fuzzy-PID control, comprising: Step 1: When the water pH and O 2 When key indicators such as temperature and water quality fluctuate significantly at the same time, the system receives the output of multi-source sensors and performs error correction and redundancy detection on the information of each channel through weighted filtering to generate an environmental fusion data vector. Based on this vector, an adaptive prediction model is built to identify extreme water quality changes and output future short-term environmental trends for subsequent deployment and lighting strategy formulation, thereby issuing early warnings at the stage of water quality mutation; The step 1 includes the following contents: Step 101: Multi-sensor data acquisition and fusion The sensor array deployed in the aquaculture water body monitors environmental parameters including pH, dissolved oxygen, 2 , water temperature T, and light intensity High-frequency data collection is performed, and all measurement readings are uploaded to the information processing unit; in order to enhance the accurate recognition of external mutations (such as sudden climate change, fluctuations in inflow and outflow, etc.), redundant detection and error correction are first performed on multi-source sensor data, and then a custom fusion formula is used to form a fusion output ,The fusion formula is based on an exponential decay mechanism for the differences between sensors, which is used to introduce a suppression factor between readings with large differences, thereby improving the reliability of the overall measurement results; Specifically, suppose at time t we have The valid observation values ​​sent back by the sensors are: , Each of these It can come from different types of sensors (such as pH sensors, dissolved oxygen sensors, etc.). First, the readings of different sensors are dimensionless or normalized to define the fusion output as follows:

[0021] Where: is the basic weight coefficient of the sensor, is the exponential amplification factor, is the difference suppression factor, and its value is greater than 0; and : represents the distance between the maximum and minimum values ​​of all sensor readings at time t, which is used to measure whether the readings are too discrete; it is an absolute value operator, used to calculate the magnitude of the numerical difference; The above fusion processing is performed on each of the environmental condition parameters respectively to obtain the fused data Rh:

[0022] The fusion results will be unified into the dynamic environment model in the next step; When in use, in a multi-sensor environment, the use of difference suppression and weighted processing can effectively filter out abnormal data points and improve the reliability of the final perception results; when the external environment changes drastically, the exponential decay term avoids abnormal readings from having excessive impact on the overall results, ensuring the stability of subsequent decisions. An exponential mechanism for suppressing differences between multiple sensors is introduced to avoid being overly sensitive to outliers when only simple weighted averaging is used.

[0023] Step 102: Dynamic environment modeling After obtaining the output fusion data, the dynamic environment model is constructed using the adaptive prediction model , using deep neural networks or fuzzy neural networks to predict short-term environmental parameters such as pH, dissolved oxygen, 2 , water temperature T, and light intensity To characterize the complex situation of the mutual coupling of multiple variables in the water environment, the historical data of each environmental condition parameter and its fusion value are input through multiple channels, and the model output can be subdivided into:

[0024] in: The prediction step length (e.g. 30 minutes or 1 hour, etc.) is the core of the dynamic environment model and will be continuously compared with the real-time collected data for continuous updating and iteration. At time t, the fused data Rh and the fused values ​​of several past moments are input into the trained or deployed neural network to obtain the multi-parameter environmental prediction information for the next period; after each prediction is completed, the prediction results are stored in the environmental prediction database and the environmental state model is generated. , used to describe the key indicator trends in a short period of time in the future. If it is monitored that certain parameters are about to exceed the safety threshold, early intervention will be carried out; Through the deep feature extraction capabilities of machine learning models or fuzzy neural networks, potential abnormal extreme weather or water quality fluctuation trends can be identified more accurately, providing a time window for preparing for the release of photosensitizers or the start-up of auxiliary equipment in advance; the dynamic environmental model continuously conducts rolling forecasts as a new batch of sensor fusion data arrives to ensure the ability to continuously track water changes; once the prediction results show that indicators such as pH or dissolved oxygen will tend to the critical range, more refined release amounts and light intensities can be formulated to reduce potential damage risks to aquatic animals.

[0025] Step 2: Complete the generation and output of environmental fusion data vectors, compare the optimal activity range of photosensitizers and the distribution of pathogens, use adaptive formula delivery technology to calculate the amount of photosensitizers delivered, and formulate a lighting plan based on the environmental matching coefficient and water body dynamic prediction results to strengthen irradiation at local hot spots and reduce energy consumption in normal areas. Through the coupling of delivery amount, delivery time, and irradiation, maximize the inactivation efficiency and reduce surface damage of aquatic animals, and avoid the accumulation of reactive oxygen species (ROS) caused by excessive light or excessive photosensitizers. The step 2 includes the following contents: Step 201: Adaptive photosensitizer delivery Based on the environmental prediction information obtained, the precise amount and timing of photosensitizer delivery are determined so that the photosensitizer PS can produce an ideal inactivation effect in the subsequent illumination stage, in order to achieve adaptive delivery: Define the basic dosage of photosensitizer PS and environmental sensitivity parameters , used to measure the extent to which aquaculture water deviates from the optimal activity range at present and in the short term; When it is predicted that environmental parameters such as pH, dissolved oxygen or temperature will deviate significantly from the ideal range of the photosensitizer, the dosage will be appropriately reduced or the administration time will be delayed to avoid excessive production of reactive oxygen species in harsh environments: Introducing environmental coordination factor , used to represent the future The suitability of the water body for the photosensitizer after a certain period of time. This coefficient is related to the prediction deviation of the first three main indicators (pH, dissolved oxygen, temperature) and is defined as follows: In the time interval Three main prediction indicators of water bodies are considered: , , ; in For any time in the interval, For the prediction step, the ideal reference value is preset separately , , , and dimensionless or normalize the different sensor readings; define the environmental deviation vector And the sensitivity weight matrix W is as follows:

[0026] This vector is used to simultaneously characterize the instantaneous deviation of the three indicators of pH, dissolved oxygen and temperature from their ideal values;

[0027] Among them, the sensitivity parameter , , The amplification weights of the criticality and deviation of pH, dissolved oxygen and temperature to the aquaculture environment are measured respectively, and the values ​​are usually positive; Using the Euclidean norm ( norm) measures the weighted environmental deviation and Integrate and accumulate it to obtain the environment deviation accumulation function :

[0028] in:

[0029] Based on the deviation cumulative function, the environmental coordination coefficient is defined as for:

[0030] Where: is a three-dimensional deviation vector, which reflects the instantaneous difference of pH, dissolved oxygen and temperature relative to the ideal value; W is a diagonal matrix, and the main diagonal elements are , , , which can be adjusted according to the breeding species and environmental characteristics; is the Euclidean norm; is the integral of the weighted deviation within the prediction interval; In order to obtain the environmental matching coefficient Then, calculate the amount of photosensitizer , the formula is: Where: is the basic dosage, which means the default dosage of photosensitizer PS under ideal conditions; Used to Map to interval, so as to make nonlinear correction to the basic delivery amount; When the environmental matching coefficient When it is too small (indicating a poor future environment), this item will significantly reduce the final delivery amount, or in extreme cases reduce it to an extremely low value; Complete photosensitizer dosage After calculation, the corresponding delivery is carried out according to the working mode of the delivery equipment (such as an adjustable pump or a quantitative feeder). If it is too low, the release time should be appropriately delayed based on actual breeding conditions, or auxiliary equipment (such as oxygenation or pH adjustment) should be used to correct the water quality before release.

[0031] When using, by introducing the environmental matching coefficient , the photosensitizer dosage and timing of delivery can be automatically fine-tuned according to the short-term predicted environmental quality, reducing waste and environmental risks. When environmental indicators are extremely abnormal, delivery will be actively reduced or postponed to avoid excessive ROS in unsuitable environments; the photosensitizer concentration after delivery is more likely to be in the optimal range, and subsequent lighting parameters can be more efficiently formulated and pathogen inactivation can be triggered, so that the photosensitizer delivery has adaptive attenuation characteristics when facing large deviations, which is far superior to the traditional fixed amount addition method. The short-term prediction data is combined with the nonlinear exponential model to avoid excessive or untimely delivery caused by relying solely on threshold judgment; Step 202: Lighting strategy formulation After completing the photosensitizer dosage After the determination and execution of and other auxiliary indicators, and then formulate a lighting strategy; the lighting strategy must ensure the matching degree between the photosensitizer and the light (wavelength, intensity, and irradiation duration), avoid continuous excessive irradiation that causes heat stress or light stress in aquatic animals, and carry out refined irradiation of the identified local hot spots, and strengthen irradiation of areas with high pathogen concentrations through mobile light sources or fiber optic arrays; Define lighting optimization function , used to determine the irradiation intensity The duration of exposure After the function is executed, the actual distribution state of the photosensitizer is used as the input variable, and the environmental matching coefficient is combined with the optimal matching relationship between the two. , which can be described as:

[0032] Where: is the base coefficient of the lighting strategy; is the amount of photosensitizer added; , Photosensitizer and The sensitivity magnification coefficient of is always a positive number; Light intensity and duration of light exposure Functions can be optimized based on lighting Obtained under the following scheduling principles: When the lighting optimization function When it is higher than the expected value (indicating that the photosensitizer PS dosage and the environmental compatibility are both good), a higher light intensity is automatically selected. Prolong the duration of light exposure , eliminate pathogens as quickly as possible within a controllable range; When the lighting optimization function When it is not higher than the expected value, it tends to reduce the light intensity And appropriately shorten the duration of light exposure , to avoid excessive irradiation causing additional stress to aquatic animals, or wasteful irradiation when the photosensitizer concentration is insufficient; This lighting scheduling can be solved by offline or online optimization algorithms (such as greedy algorithms, genetic algorithms, etc.), and can also be fine-tuned by a microprocessor in the subsequent real-time closed-loop stage.

[0033] When implementing actual irradiation, spatial zoning will be carried out according to the distribution of pathogens: The hotspot area is illuminated with relatively high intensity, and the non-hotspot area is illuminated with normal intensity or no illumination. The time interval between the release and illumination is determined by the illumination optimization function The photosensitizer concentration reflected in the formula is determined by the degree of coordination; when the expected photosensitizer is fully diffused in the water body and has not significantly attenuated, the illumination can be started in a timely manner; the pathogen concentration or related indicators of each area are obtained in real time through sensors (such as fluorescence imaging, microbial counting or other rapid detection methods) and compared with the preset threshold. When the detection value of a certain area exceeds the threshold, the area is defined as a hot spot.

[0034] When used, it realizes the coordination of delivery and irradiation. and Key quantities such as light intensity and light intensity are directly involved in lighting decisions, avoiding the waste or inefficiency caused by the traditional separate control of photosensitizers and irradiation; reducing environmental and biological stress, achieving dynamic inhibition of excessive irradiation by appropriately increasing or shortening the lighting duration, and making preventive adjustments to the tolerance threshold of aquatic animals to improve the efficiency of pathogen killing: forming a differentiated strategy between enhanced irradiation in hot spots and conventional irradiation in general areas, taking into account both comprehensive killing and key strikes.

[0035] Through the linkage between adaptive photosensitizer delivery and lighting strategy formulation, multi-dimensional environmental coordination coefficient It can not only accurately schedule the release amount, but also be integrated with the distribution of photosensitizers into the lighting strategy to produce a strong synergistic effect. This synergistic mechanism enables the photosensitizer and the lighting link to leverage and check each other, maintaining a high degree of effectiveness against pathogens while reducing stress on aquatic animals and waste of resources.

[0036] Step 3: After the photosensitizer has been placed and illumination has begun, if the sensor detects that the environmental condition data deviates significantly from the target again, the mixed fuzzy-PID real-time closed-loop adjustment is performed, and the scene difference vector is used to adapt to the water body regulation needs. When the water environment fluctuates violently or the risk of pathogen spread increases, the local irradiation is strengthened or the isolation device is activated in time. After each control cycle, the deviation vector and compensation instruction are stored as data; The step three includes the following contents: Step 301: Real-time closed-loop control Based on the results of photosensitizer placement and light parameter setting, and comparing the real-time sensor data of the actual environment, if the current state is observed to be consistent with the set target range (in or As a reference, if there is a significant deviation, the closed-loop correction mode will be immediately entered. This mode is executed by a microprocessor (such as STM32 or other high-performance embedded controllers). The core logic is as follows: Get new environmental measurements at time t to form an instant state vector :

[0037] Compared with the previous stage Or the ideal interval vector expected by the lighting strategy Compare and get the environmental deviation vector :

[0038] A control method based on nonlinear potential function is used to transform the environmental deviation vector Mapping to control signal vector , including compensation for light intensity and start and stop scheduling of auxiliary devices , which can be written as:

[0039] in, It is a nonlinear potential function, which is used to amplify or suppress the size of the deviation vector in sections; is a mapping operation used to convert the potential function output into a command value that can directly drive the actuator. For example, when the absolute value of the deviation is small, The signal can be amplified appropriately to achieve agile adjustments; when the deviation is extremely large, overly drastic corrections can be limited within safety limits to avoid secondary stress on aquatic animals; Light compensation: If light intensity compensation If it is positive, the light source power is increased; if it is negative, the light source power is reduced or the effective irradiation time is shortened ; Auxiliary device scheduling: auxiliary device start and stop scheduling It can trigger the oxygenation equipment, pH buffering agent delivery unit, local water flow isolation device, etc., and adaptively control according to the source of deviation. For example, if the pH is too low, the buffering adjustment will be triggered; if the dissolved oxygen is insufficient, the oxygenation equipment will be started, etc. When in use, through the above closed-loop control function, the environmental parameter error is corrected nonlinearly and quickly in each control cycle, which not only improves the immediate efficiency of photodynamic inactivation, but also reduces the risk of rapid spread of pathogens due to local deterioration of the environment; close connection with the delivery and lighting strategy: all correction actions revolve around the ideal interval vector To achieve the adaptive linkage of delivery-irradiation-control; introduce nonlinear potential function It can handle the deviation, which greatly exceeds the conventional linear PID or the method based on variance detection only, and has higher response flexibility.

[0040] Step 302: Multi-scenario collaborative scheduling Further, in view of the differences in characteristics of different scenarios such as indoor high-density farming, outdoor pond farming and caged floating water farming, a cross-scenario scheduling strategy is implemented. The core of this strategy is: according to the previously determined light pattern and photosensitizer distribution plan, combined with the new round of sensor data and closed-loop control instructions, coordinated scheduling is carried out in spatial and temporal dimensions to ensure that the ideal inactivation effect and aquatic animal safety can be achieved in each scenario; Define the environment difference vector , whose components may include specific scene factors such as water flow velocity, aquaculture density, light interference, etc. For example:

[0041] in: Represents the water flow rate (larger in cage culture), Represents the breeding density (larger in indoor high-density systems), Represents the degree of external light interference (more obvious in outdoor ponds); Build multi-scenario scheduling functions , the control signal vector Difference vector with environment Perform joint mapping to obtain scenario-based instructions :

[0042] When the water flow rate When it is larger (i.e. cage water area), multi-scenario scheduling function Compensation for light intensity Do partition enlargement to compensate for the decrease in photosensitizer concentration caused by water dilution; When the stocking density When the value is large (indoor high-density scene), multi-scene scheduling function Auxiliary device start and stop scheduling can be called for auxiliary devices Prioritize enhancement to prevent localized outbreaks in densely populated areas where pathogens can spread quickly; If the light interference Too high (outdoor pond is strongly affected by sunlight), multi-scene scheduling function Automatically suppress excessive increase in light intensity and extend lighting time , to reduce the risk of overheating or light burns to aquatic animals; Scenario-based instructions Finally, the scheduling module sends it to each independent execution subsystem for execution: Mobile lighting devices: track position and adjust irradiation power according to the local water flow rate of the cage or pond; local enclosure / isolation devices: activate temporary surround networks in high-flow scenarios to reduce the spread of pathogens; oxygenation and water quality regulators: prioritize more power and resources in high-density scenarios; external light compensation / shielding equipment: add sunshades or reflective devices if outdoor interference is too strong; When used, the environment difference vector The introduction of the same closed-loop control strategy enables differentiated execution plans based on the core characteristics of the actual farming scenario, combined with the control signal vector Difference vector with environment , can realize dynamic allocation for different areas and time periods, enhance resource utilization and inactivation accuracy; the same control signal vector Just add a layer of multi-scene scheduling function on the outer layer It can adapt to different scenarios and reduce the need for system hardware modification; multi-scenario scheduling function Multiple factors such as flow rate, density, and light interference are uniformly processed in vector form, no longer relying on simple empirical thresholds but achieving more delicate control allocation through a mapping mechanism.

[0043] Step 4: When the photodynamic inactivation process ends or reaches the expected duration, the treatment effect is summarized and evaluated based on the aforementioned compensation instructions and control feedback indicators, and the replay pool mechanism in deep reinforcement learning is used to train each state-action-reward sequence, update the adaptive delivery and lighting strategy model, and combine the data curve generated in the evaluation process with the scene difference vector to summarize the key environmental influencing factors and solidify the optimized parameters into the knowledge base; The step 4 includes the following contents: Step 401: Multi-dimensional effect evaluation Based on the sensor records, control instruction records and final water pathogen inactivation results generated during the actual implementation process, and against the set target indicators, a systematic multi-dimensional evaluation is conducted, mainly including the following contents: Record pathogen clearance rate through fluorescent imaging device or other pathogen detection module ; Record individual injury rates by observing and testing surface injuries or other health indicators of aquatic animals ; Record data on oxygen enrichment equipment, isolation devices, light source power, etc. to obtain comprehensive energy consumption indicators , characterizes the total energy consumption of photosensitizer placement, illumination, and various auxiliary equipment; And monitor after treatment to Recovery of pH, dissolved oxygen, temperature, etc. within a certain period of time; If the water quality can quickly return to the safe range, and the comprehensive energy consumption index If it remains at a relatively low level, it means that this round of governance is more efficient in terms of energy conservation and ecological balance; based on the above three types of evaluation results, an effect evaluation report is formed and output; When in use, the pathogen inactivation rate and aquatic animal safety can be taken into account while paying attention to energy consumption and water quality recovery to achieve a comprehensive evaluation of the entire treatment process.

[0044] Step 402: Deep self-learning optimization Using the output effect evaluation report and the recorded execution actions, an adaptive optimization mechanism is constructed. The mechanism is based on the state-action-reward logic cycle to provide control and improvement for each round of photodynamic inactivation process, including: Defining states , which may include:

[0045] Multiple environment difference vectors can also be added etc., to ensure that the complete environment and operation characteristics are reflected in the reinforcement learning model; Defining Actions : action Refers to the adjustment of photosensitizer placement and lighting parameters, such as changing the amount of photosensitizer placed Correction value, fine-tuning light intensity compensation value And the light duration compensation value , or control the start and stop of auxiliary devices, etc. All executed control instructions can be recorded as an action instance.

[0046] To balance pathogen inactivation efficiency with aquatic animal safety and energy consumption, a comprehensive reward function is defined , extract the key points from the evaluation metrics:

[0047] Where: , , is the balance coefficient, which is a positive number ranging from 0.1 to 10 and is used to adjust the relative importance of the three indicators in the reward function; , , is the nonlinear amplification factor, which is a positive number and can be between 0.5 and 5; The larger the value, the better the balance between inactivation efficiency, aquatic animal safety and energy consumption. Otherwise, corresponding strategy corrections must be made in the next round of training. The execution sequence of each cycle Stored in the replay pool, after iterative training through Q-learning or deep Q network (DQN), policy gradient and other methods, the mapping strategy is finally formed: ; When used, through the comprehensive reward function It can find a better compromise between pathogen elimination, animal safety, and energy consumption to avoid local optimality. The new data after each round of governance will participate in the next round of training. The algorithm will continue to evolve and constantly correct unreasonable delivery or irradiation methods. The environmental difference vector A reinforcement learning model is also incorporated to adapt the strategy to multiple scenarios such as high-density indoors, outdoor water bodies, and cage mobility.

[0048] Step 403: Knowledge base update and next batch application preconfiguration The strategy model parameters and key evaluation data obtained from this round of training are stored in the knowledge base for direct use in the next batch or cycle of farming practices. The output of the reinforcement learning model (or neural network weights, Q tables, etc.) is stored separately according to different scene labels (such as indoor, outdoor, cages), and the mapping relationship of each key variable is retained to facilitate rapid matching of similar scenes. For situations where manual or expert intervention is still required in certain extreme situations (such as severe dissolved oxygen deficiency and extreme temperature deviation), expert rules (such as limit thresholds or forced start-stop strategies) are allowed to be used in parallel with reinforcement learning models and stored in the same knowledge base. If similar extreme conditions occur, expert rules can be triggered first to prevent serious accidents. Before a new batch of breeding begins, the corresponding strategy parameter set is automatically loaded according to the scene label, and the initial photosensitizer dosage is preset. , Environmental coordination coefficient Thresholds and lighting optimization plans, if there are major environmental changes, additional fine-tuning will be initiated; When in use, each governance process will produce new empirical data and reinforcement learning model results, the knowledge base will be continuously upgraded, and the overall system capabilities will be accumulated and improved over time. There is no need to build a model from scratch before each start, and the optimal strategy in similar scenarios can be directly imported, greatly shortening the preparation cycle. In the case of data scarcity or extreme abnormalities, expert rules can complement the self-learning model to ensure security and controllability.

[0049] Step 5: For large-scale farming scenarios with many partitions or across regions, local closed-loop scheduling is achieved by deploying edge control units in each sub-region, and the environment fusion data and control instruction logs are uploaded to the cloud regularly. Multi-objective training is performed using deep learning algorithms to form a general decision-making model, which is then sent to each edge control unit for synchronous execution; The step five includes the following contents: Step 501: Multi-node hierarchical control architecture When the scale of farming is large, the areas are multi-divided or the geographical locations are discrete, the real-time feedback speed and load capacity of a single control center may not be able to meet the high-frequency decision-making needs. A multi-node hierarchical control architecture is proposed. Specifically, an edge control unit is deployed in each sub-division (or local farming unit) to quickly perform the following operations in a local area: Each edge control unit receives real-time environmental condition data (pH, O 2 ,T, etc.), and use the updated expert knowledge base or reinforcement learning strategy to adjust the photosensitizer dosage , irradiation intensity The duration of exposure Equivalent feedback control strategies for fast adaptive adjustment at the edge level; If the environmental coordination factor Or if there is a local extreme deviation in the distribution of pathogens in a sub-partition, the edge control unit can respond quickly under low latency conditions to avoid the delay caused by remote commands from the main control center; When multiple partitions are adjacent, the edge control units communicate key status information (such as whether the oxygenation equipment is started, whether the pathogen has spread in the adjacent partition, etc.), and coordinate the implementation of cross-partition isolation or water flow regulation within the partition: This coordination is usually achieved through a lightweight message exchange protocol (such as MQTT or a custom protocol); Each edge control unit makes a first-level rapid decision in the local environment and transmits the data summarized in stages (including the local inactivation rate , damage rate , Energy consumption The main control center will fuse the multi-zone information and make more macro-level decisions, such as uniformly allocating the upper limit of light source power, placing photosensitizer rations, or comparing water quality differences between different administrative regions.

[0050] When in use, the edge control unit performs high-frequency closed-loop adjustment locally to reduce delays or network instability caused by long-distance communications. Failure of the edge control unit in any partition will not immediately paralyze the entire system. The main control center can dispatch other partitions to coordinate or implement emergency mode.

[0051] The first level of rapid decision-making means that after each edge control unit (ECU) collects sensor data locally in real time, it immediately analyzes and compares local water quality parameters (such as pH, dissolved oxygen, temperature, etc.) based on the preset closed-loop control strategy and fuzzy-PID algorithm, and then quickly adjusts the light intensity, photosensitizer dosage, or activates the operation of auxiliary equipment (such as oxygen pumps, isolation devices). This decision-making process is completed within milliseconds to seconds, aiming to respond quickly to abnormal changes in the local environment, thereby ensuring the overall stability of the system and real-time control performance without waiting for central control or cloud feedback.

[0052] Step 502: Cloud-based big data training platform The data of different sub-districts or farms are centrally processed and trained with deep learning to obtain a more universal inactivation model and strategy, which greatly reduces the local training cost of each edge control unit and improves accuracy when it is subsequently issued. The key steps are as follows: Regularly report the environmental parameters, control actions and evaluation indicators (inactivation rate) generated by each edge control unit during local operation , Aquatic Animal Injury Rate , Energy consumption The model is then uploaded to the cloud big data platform (such as the cloud-based big data platform), and auxiliary information such as geographic location, aquatic biological data, and seasonal temperature characteristics are added, so that the model training process can capture a wider range of environmental differences; Construct a multi-objective optimization deep learning framework on the cloud platform, including a multi-task network structure or a multi-head output layer, corresponding to sub-objectives such as inactivation efficiency, aquatic biological safety, and energy consumption control; Based on the following exemplary generalization function , perform large-scale batch training on the collected data:

[0053] Where: , , The difference between the true value and the model prediction value in terms of inactivation rate, damage rate, and energy consumption is measured respectively (it can be a high-order loss metric, or a more advanced nonlinear metric based on logarithmic mapping, etc.), and the value is between 0 and 1; , , Adjustable target weights support dynamic priority allocation in different application scenarios , , It is the model prediction output, which is used to compare with the true value and transfer the gradient in reverse. The recommended value range is between 0.1 and 10. After training is completed, the cloud will use the new cloud decision model Sent to each edge control unit, where the cloud decision model It refers to a general decision model obtained by deep learning (such as reinforcement learning, deep Q network or policy gradient algorithm) training on a cloud big data training platform by integrating environmental monitoring data, control logs and evaluation indicators from each edge control unit (ECU); if the breeding environment of a sub-division or region is highly matched with the cloud training data, the cloud decision model can be directly used Achieve high prediction accuracy; if there are still local differences, secondary fine-tuning can be performed locally in the edge control unit based on the reinforcement learning mechanism; When in use, large-scale data from multiple regions, multiple species, and multiple seasons are collected, and the trained model has better generalization capabilities for new scenarios, shortening local exploration and trial and error time; cloud-based training can utilize high-performance computing resources to process massive data in parallel, reducing the hardware requirements for each edge control unit; as the system deployment scope expands, the cloud platform can regularly retrain and iterate the model to quickly absorb the latest aquaculture and pathogen prevention and control information globally or nationwide.

[0054] Step 503: Cross-scenario data fusion and global strategy iteration Cloud-based decision model Combined with the data from the partitioned edge control units, cross-scenario data fusion and global strategy iteration are performed for different farming scenarios, further ensuring that various differentiated scenarios such as indoor high-density, outdoor ponds and cages can be flexibly responded to and the global optimal strategy can be regularly formed and fed back to the self-learning process; In the proposed environmental difference vector The environmental difference vector reported by different edge control units is analyzed by quantifying factors such as flow rate, breeding density, light interference, etc. Perform unified processing and compare with the cloud model to identify possible blind spots of scene differences, add new feature dimensions or modify existing feature weights; The above scenario difference information is combined with the generated cloud decision model Combination, using distributed strategy iteration algorithm:

[0055] Where: It is a distributed fusion function used to aggregate the feedback results of each edge control unit and then perform the cloud decision model Fine-tune and generate a global decision model ; In the above formula Indicates The data provided by each sub-partition or sub-scenario corresponds to the indicators obtained through local execution or self-learning.

[0056] Global decision model after iteration It has the comprehensive advantages of taking into account the differences in different scenarios and still maintaining local optimal characteristics based on big data training.

[0057] Finally, the global decision model The optimization results extracted will flow back into the reinforcement learning or knowledge base update process, so that subsequent batches can also use distributed global experience in a single scenario, accelerate adaptation and reduce exploration costs.

[0058] When in use, the rapid feedback capability of high-density indoor scenes, the rich water quality change data of outdoor water bodies, and the mobility characteristics of cage scenes are integrated to obtain a more comprehensive global decision; each new aquaculture partition can be systematically included in the distributed architecture, and its data can provide more samples for cloud model training and partition self-learning. Using distributed strategy to iterate distributed fusion function Implementing a cloud-based decision model The organic integration of the local reinforcement learning results of each partition edge control unit breaks through the limitations of traditional single-center or single-node learning and introduces the environment difference vector A unified management mechanism allows breeding data in different ecological environments to be cross-verified and enhanced.

[0059] To ensure the multi-sensor data fusion and environmental coordination coefficient in the present invention , photosensitizer dosage And lighting parameters In order to keep the physical dimensions consistent during the calculation process, it is hereby declared that the different parameters involved in the relevant formulas or functions, such as pH, dissolved oxygen mg / L, temperature ℃, light intensity W / m², etc., must be converted or normalized before use, or unified by introducing weight coefficients containing unit adjustment factors. Specifically, if the pH difference, dissolved oxygen deviation, and temperature deviation are superimposed, exponentially operated, or multiplied in the same expression, they are respectively subtracted from the corresponding ideal values ​​and then divided by their respective predetermined characteristic scales or normalization factors, so as to map the variables of different dimensions to comparable dimensionless intervals; in addition, in fuzzy-PID control or environmental fusion calculations, the corresponding dimension matching of each sensor item is also required, so that the calculation model disclosed in the present invention has a clear and reproducible physical meaning in actual implementation, thereby avoiding the problem of insufficient disclosure due to inconsistent dimensions.

[0060] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0061] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0062] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only some logical function divisions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0063] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0064] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for adjusting photodynamic inactivation parameters based on hybrid fuzzy-PID control, characterized in that: include, When environmental condition parameters exceed the preset fluctuation range, the multi-sensor readings are integrated and the errors are corrected, and an adaptive prediction model is constructed to convert the environmental state from multi-source discrete quantities to short-term trends to warn of extreme changes; When the environmental model shows that the water quality index is deviating outside the threshold, the photosensitizer dosage is calculated using an adaptive formula based on the environmental matching coefficient, and the illumination parameters are configured through a hybrid fuzzy-PID algorithm to match the photosensitizer injection rhythm with the light source output, switching the static estimation to adaptive drive. If the real-time sensor deviates from the target range set by the reading, the hybrid fuzzy-PID closed loop is called to adjust the lighting parameters and auxiliary equipment instructions, and the oxygenation or isolation action is determined in combination with the scene difference vector to restore the local water quality from the unstable state to the safe range; After the photodynamic inactivation phase ends or reaches the time limit, the control feedback data and control signal sequence are summarized and input into the reinforcement learning module to perform replay pool training and update the control strategy parameters, so that the single operation data is transformed into a continuously iterative self-learning knowledge base; When the strategy update is completed and a large-scale cross-partition scenario needs to be faced, each edge control unit receives the cloud decision model and combines the environmental difference vector to carry out multi-node hierarchical scheduling, so that the original single-point closed loop is expanded to cloud-edge collaboration, and the distributed environmental differences are integrated into a global strategy.

2. The method for adjusting photodynamic inactivation parameters according to claim 1, characterized in that: The sensor array deployed in the water collects environmental condition parameters at a high frequency, performs redundant detection and error correction, and then uses a custom fusion formula to form a fusion output. The fusion formula is based on the exponential decay mechanism of the differences between sensors.

3. The method for adjusting photodynamic inactivation parameters according to claim 2, characterized in that: Use adaptive prediction models to build dynamic environmental models, and use deep neural networks or fuzzy neural networks to predict the trend of environmental condition parameters in the short term to obtain multi-parameter environmental prediction information for the next period; The prediction results are stored in the environmental prediction database and an environmental status model is generated. If it is monitored that certain parameters are about to exceed the safety threshold, early intervention is carried out.

4. The method for adjusting photodynamic inactivation parameters according to claim 3, characterized in that: Based on the environmental prediction information obtained, the environmental matching coefficient is introduced to characterize the suitability of the water body for the photosensitizer delivery, and the photosensitizer delivery amount is calculated and obtained, and the corresponding delivery is implemented according to the working mode of the delivery equipment; If the environmental coordination coefficient is lower than expected, the release time will be adaptively delayed, or auxiliary equipment will be called in to correct the water quality before release.

5. The method for adjusting photodynamic inactivation parameters according to claim 4, characterized in that: After the photosensitizer dosage is determined and executed, the lighting strategy is formulated based on the concentration distribution of the photosensitizer in the water body at the current moment and the prediction results. The lighting strategy is solved by an offline or online optimization algorithm. The refined irradiation of local hot spots enables the optimal linkage between the photosensitizer concentration distribution and the light source output in the time and space domain.

6. The method for adjusting photodynamic inactivation parameters according to claim 1, characterized in that: Relying on the photosensitizer delivery and light parameter settings, and comparing the real-time sensor data of the actual environment, if the current state is observed to deviate significantly from the set target range, it will enter the closed-loop correction mode; A control method based on nonlinear potential function is adopted to map the environmental deviation vector to the control signal vector, including compensation for light intensity and start-stop scheduling of auxiliary devices, and adaptive control is performed according to the source of the deviation.

7. The method for adjusting photodynamic inactivation parameters according to claim 6, characterized in that: According to the previously determined illumination pattern and photosensitizer distribution plan, combined with the new round of sensor data and closed-loop control instructions, coordinated scheduling is carried out in spatial and temporal dimensions to ensure that the ideal inactivation effect and aquatic animal safety can be achieved in each scenario; Define the environmental difference vector, whose components include water flow velocity, aquaculture density, and light interference degree; Construct a multi-scenario scheduling function that jointly maps the control signal vector and the environment difference vector to obtain scenario-based instructions, and send them to each independent execution sub-unit for execution.

8. The method for adjusting photodynamic inactivation parameters according to claim 1, characterized in that: Based on the sensor records, control instruction records and execution feedback data generated during the actual execution process, a multi-dimensional evaluation is conducted against the set target indicators, and an effect evaluation report is generated and output. In addition, an adaptive optimization mechanism is constructed in combination with the recorded execution actions. The adaptive optimization mechanism is based on a state-action-reward logic loop, where: Define states, actions, and comprehensive reward functions; make corresponding strategy corrections and optimizations during training based on the comprehensive reward function, store the execution sequence of each cycle in the reinforcement learning replay pool, and form a mapping strategy after iterative training.

9. The method for adjusting photodynamic inactivation parameters according to claim 8, characterized in that: The trained policy model parameters and key evaluation data are stored in the knowledge base, the reinforcement learning model output is stored separately according to different scenario labels, and the mapping relationship of each key variable is retained; Expert rules are used in parallel with reinforcement learning models. Expert rules are triggered first when preset rules are met, and the corresponding strategy parameter set is automatically loaded according to the scene label. If major environmental changes occur, supplementary fine-tuning is initiated.

10. The method for adjusting photodynamic inactivation parameters according to claim 1, characterized in that: An edge control unit is deployed in each sub-division to collect water quality dynamics within the sub-division and perform hybrid fuzzy-PID closed-loop adaptive regulation. When multiple partitions are adjacent, the edge control units communicate key status information and coordinate the implementation of cross-division isolation or water flow regulation within the partition. After making a decision in the local environment, the edge control unit transmits the phased summarized data to the main control center, which then integrates the multi-partition information.

11. The method for adjusting photodynamic inactivation parameters according to claim 10, characterized in that: Regularly upload the environmental parameters, control actions, and evaluation indicators generated by each edge control unit during local operation to the cloud big data platform and attach auxiliary information; After building a multi-objective optimization deep learning framework on the cloud platform, large-scale batch training is performed on the collected data based on the generalization function. After the training is completed, the new cloud decision model is sent to each edge control unit.

12. The method for adjusting photodynamic inactivation parameters according to claim 10, characterized in that: If the farming environment of a certain sub-district or region is highly matched with the cloud training data, the cloud decision model can be directly adopted; if there are still local differences, secondary fine-tuning is performed locally in the edge control unit based on the reinforcement learning mechanism; Based on the cloud-based decision-making model and partitioned edge control unit data, cross-scenario data fusion and global strategy iteration are performed for different breeding scenarios, and the global optimal strategy is regularly formed and fed back to the self-learning process.

13. The method for adjusting photodynamic inactivation parameters according to claim 12, characterized in that: The environmental difference vectors reported by different edge control units are processed uniformly and compared with the cloud model to identify possible scene difference blind spots, add new feature dimensions or modify existing feature weights; The scene difference information is combined with the generated cloud-based decision model, and the distributed policy iteration algorithm is used to fine-tune the cloud-based decision model after summarizing the feedback results of each edge control unit to generate a global decision model.

Images