A multi-factor combined regulation system based on an aquaculture factory

Abstract

The present application relates to the technical field of aquaculture factory regulation, and particularly relates to a multi-factor combined regulation system based on an aquaculture factory, comprising: a data acquisition module, used for acquiring multi-dimensional factor data of water quality, biology, equipment and environment in the aquaculture factory, wherein the multi-dimensional factor data at least includes water quality factor data and biological factor data. The present application collects multi-dimensional factor data of water quality, biology, equipment and environment, and after preprocessing of smoothing denoising, outlier rejection and missing value completion and feature extraction, a multi-factor interaction model containing a water quality dynamic prediction sub-model, a growth prediction sub-model and a biomass estimation sub-model is constructed, the interaction law of the four sub-models is analyzed, and then a water quality, feeding, equipment coordination and energy consumption regulation strategy is generated in combination with a preset breeding target. The present application can also self-learn and adjust model parameters through reinforcement learning, a precise driving device is driven by an execution module to execute a regulation action, and a feedback optimization module monitors the effect and generates a correction instruction.

Description

Technical Field

[0001] This invention relates to the field of aquaculture factory-style regulation technology, specifically a multi-factor joint regulation system based on aquaculture factories. Background Technology

[0002] As aquaculture transforms towards intensification and intelligentization, recirculating aquaculture systems (RAS) have become the mainstream development direction in the industry. This aquaculture model achieves water recycling through artificial control of the aquaculture environment, effectively overcoming the limitations of the natural environment and improving aquaculture efficiency and stability. Its core lies in the precise control of various factors affecting aquaculture stability.

[0003] Among existing published patents, such as CN113728970B, the patent title is "Intelligent Control System and Method for Variable Speed ​​Flow in Recirculating Aquaculture Based on Multi-Parameter Joint Control." This patent monitors ammonia nitrogen, dissolved oxygen, turbidity, and water level parameters in each pool of the aquaculture water, divides the feeding interval into four stages, and adjusts the frequency of the frequency converter to control the water pump flow rate, thereby achieving water quality maintenance and energy consumption optimization.

[0004] However, existing multi-parameter control systems do not deeply couple the biological characteristics of the cultured organisms with water quality, equipment, and environmental factors. These systems focus only on controlling the physicochemical parameters of the water and the operating parameters of the equipment, neglecting the dynamic influences at the biological level, such as the real-time growth status and changes in group behavior of the cultured organisms. They cannot dynamically optimize control strategies based on fluctuations in these biological characteristics. This leads to a disconnect between the control behavior and the actual needs of the cultured organisms, insufficient control precision, and difficulty in maintaining long-term steady-state in the culture process, thus affecting the survival rate and economic benefits of the cultured organisms. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-factor joint regulation system based on aquaculture farms, which solves the problem that existing technologies fail to deeply couple the biological characteristics of aquaculture organisms with water quality, equipment, and environment, resulting in a disconnect between regulation and actual needs and difficulty in maintaining long-term aquaculture stability.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-factor joint control system based on an aquaculture farm, comprising: The data acquisition module is used to collect multi-dimensional factor data that affect the aquaculture homeostasis within the aquaculture factory. The multi-dimensional factor data includes at least water quality factor data, biological factor data, equipment factor data, and environmental factor data. The data processing and modeling module is used to receive the multi-dimensional factor data, perform data preprocessing and feature extraction, and construct a multi-factor interaction model. The multi-factor interaction model is used to analyze the interaction patterns among water quality, organisms, equipment and environment. The multi-factor joint regulation and decision-making module is used to generate coordinated regulation instructions based on the output results of the multi-factor interaction model and the preset breeding target. An execution module is used to receive the coordinated control instructions and drive the corresponding aquaculture equipment to perform control actions; The feedback optimization module is used to monitor the execution effect of the control action, collect multi-dimensional factor data after control, and compare the multi-dimensional factor data after control with the preset breeding target to form a feedback signal for correcting the multi-factor interaction model or collaborative control command.

[0007] Furthermore, the data acquisition module is specifically used for: The water quality data, including dissolved oxygen, ammonia nitrogen, water temperature, pH value, nitrate nitrogen, and CO2 concentration, are collected by sensors deployed in aquaculture ponds and water treatment systems. The biological factor data is collected using machine vision equipment, acoustic sensors, and inspection robots. The biological factor data includes the growth status of the cultured objects, group behavior, group biomass, and physiological indicators related to disease early warning. The equipment factor data is collected through the monitoring units on the feeding equipment, water treatment equipment, inspection equipment and cleaning equipment. The equipment factor data includes the operating parameters and status data of each piece of equipment. The environmental data, including workshop temperature and humidity, light intensity, and atmospheric pressure, are collected by sensors deployed in the breeding workshop.

[0008] Furthermore, the data processing and modeling module, when performing data preprocessing, is specifically used for: The multi-dimensional factor data is subjected to smoothing and denoising processing to eliminate noise caused by water body interference; Outlier removal was performed on the multi-dimensional factor data based on statistical principles to eliminate erroneous data caused by sensor drift. Interpolation is used to fill in the missing values ​​in the multi-dimensional factor data to generate a complete standardized dataset.

[0009] Furthermore, the data processing and modeling module, when performing feature extraction, is specifically used for: The dynamic change characteristics of water quality indicators are extracted from the water quality factor data, including the diurnal fluctuation range of dissolved oxygen. Biological behavioral characteristics are extracted from the biological factor data, including feeding evenness and frequency of abnormal behavior; The equipment operating characteristics are extracted from the equipment factor data, including the correlation curve between the oxygenator's energy consumption and dissolved oxygen enhancement efficiency.

[0010] Furthermore, when constructing a multi-factor interaction model, the data processing and modeling module is also used to construct the following sub-models: A dynamic water quality prediction model is used to predict future trends in water quality indicators by combining factors such as feeding amount, stocking density, and water temperature. Growth prediction models are used to predict the future growth status of farmed organisms by combining feed input, water quality parameters, and aquaculture environment data. A biomass estimation model is used to fuse instance segmentation results from machine vision with the output of a density estimation network to estimate population biomass online. The multi-factor interaction model is constructed based on the outputs of the water quality dynamic prediction model, the growth prediction model, and the biomass estimation model.

[0011] Furthermore, when generating coordinated control instructions, the multi-factor joint regulation decision-making module is specifically used to generate the following strategies: A water quality control strategy is used to dynamically adjust the aerator power, water treatment system operating parameters, and water exchange flow rate based on the prediction results of the water quality dynamic prediction model and the biological needs reflected by the biological factor data. Feeding regulation strategy, used to adjust feeding amount, feeding area and feeding frequency based on the estimation results of the biomass estimation model, the prediction results of the growth prediction model and the biological behavior characteristics; Equipment coordination strategy is used to coordinate the operation sequence of feeding equipment, oxygenation equipment and cleaning equipment to achieve conflict-free operation of equipment; Energy consumption control strategies are used to adjust the operating time and power of high-energy-consuming equipment based on real-time energy consumption data and aquaculture needs.

[0012] Furthermore, the multi-factor joint regulation decision-making module is also used to perform self-learning adjustments, specifically in the following manner: Acquire historical control data and aquaculture effect feedback data provided by the feedback optimization module; Using reinforcement learning algorithms, the parameters in the water quality dynamic prediction model, growth prediction model, biomass estimation model, and multi-factor interaction model are iteratively adjusted based on the historical regulation data and aquaculture effect feedback data. The optimized model parameters will be applied to the generation of the next round of coordinated control commands.

[0013] Furthermore, the execution module includes: The water quality control execution unit is used to control the aerator, water pump, biological filter, disinfection equipment and decarbonization device to perform dissolved oxygen enhancement, ammonia nitrogen removal and water disinfection operations; The feeding execution unit is used to control the track-mounted intelligent feeding robot to perform precise feeding and zoned feeding operations; The equipment collaborative execution unit is used to control the inspection robot, the cleaning robot, and the handling robot to complete the inspection, cleaning, and material handling tasks according to the timing sequence in the collaborative control instructions. The environmental control execution unit is used to control the workshop's air conditioning and lighting equipment, adjusting the workshop's temperature, humidity, and light intensity to the set range.

[0014] Furthermore, when generating the feedback signal, the feedback optimization module is specifically used for: The assessment evaluates the alignment between multi-dimensional data after regulation and the preset aquaculture targets. By calculating the compliance rate of water quality indicators, the deviation of biological growth status, and the change rate of energy consumption, an effect evaluation index is generated. When the effect evaluation index shows that a certain factor index deviates from the preset breeding target, the cause of the deviation is analyzed and a correction instruction for adjusting the relevant control strategy is generated. Data on the regulation process, effect evaluation indicators, and correction instructions are archived in a cloud database to provide data support for model iteration.

[0015] Furthermore, the system is deployed using a three-tiered "cloud-device-edge" architecture: The edge layer, deployed at the breeding farm site, includes the field acquisition equipment, execution module, and edge computing nodes in the data acquisition module, and is used to perform real-time data acquisition and local rapid regulation; The cloud layer is equipped with a big data management and control platform, which is used to perform data storage, model training, global control decision-making, and remote monitoring. The edge layer connects the edge layer and the cloud layer, and is used to perform data preprocessing, instruction forwarding, and local collaborative control.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention collects multi-dimensional data on water quality, organisms, equipment, and the environment. After preprocessing including smoothing, noise reduction, outlier removal, and missing value completion, and feature extraction, it constructs a multi-factor interaction model containing sub-models for dynamic water quality prediction, growth prediction, and biomass estimation. The model analyzes the interaction patterns among these four factors and, combined with preset aquaculture targets, generates strategies for water quality, feeding, equipment coordination, and energy consumption control. Furthermore, it utilizes reinforcement learning to adjust model parameters, enabling the execution module to precisely drive corresponding equipment to perform control actions. The optimization module provides feedback on monitoring results and generates correction instructions. This process deeply couples the biological characteristics of the aquaculture organisms with water quality, equipment, and the environment, solving the problems of disconnect between control and the actual needs of the aquaculture organisms, insufficient precision, and difficulty in maintaining long-term steady-state conditions in existing technologies. It effectively maintains aquaculture steady-state conditions, improving survival rates and profitability. The cloud-edge three-level architecture balances real-time response and global optimization, adapting to factory-style aquaculture, reducing energy consumption control costs, and is compatible with different aquaculture species, demonstrating good versatility and practicality. Attached Figure Description

[0017] Figure 1 This is an overview diagram of the system architecture of the present invention; Figure 2 This is a flowchart of the data acquisition and preprocessing process of the present invention; Figure 3 This is a flowchart of the multi-factor modeling process of the present invention; Figure 4 This is a flowchart of the joint regulation and decision-making process of the present invention; Figure 5 This is a flowchart of the feedback optimization process of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1

[0020] Please see Figure 1-5 This embodiment combines the application scenario of factory-scale recirculating aquaculture of Litopenaeus vannamei with a detailed explanation of the specific technical details and operation process of a multi-factor joint control system based on aquaculture factories, to ensure that the system can accurately adapt to the aquaculture needs and achieve steady-state control of the aquaculture process.

[0021] First, the system is deployed using a three-tier architecture of "cloud-device-edge". This architecture can take into account both real-time control and response and global optimization decision-making, and is adapted to the large-scale and intelligent needs of factory farming. The edge layer is deployed at the aquaculture plant site. Based on the layout of 10 500m³ aquaculture ponds and supporting water treatment systems, one dissolved oxygen sensor is deployed at each of the four corners of each pond, 0.5m from the bottom. The sensor's measurement range is 0-20mg / L, with an accuracy of ±0.1mg / L. The data sampling frequency is set to once per minute, determined based on the dissolved oxygen change rate and monitoring accuracy requirements in Litopenaeus vannamei aquaculture, enabling timely capture of dynamic fluctuations in dissolved oxygen. Simultaneously, machine vision equipment using high-definition cameras with a frame rate of 30 frames per second is deployed in the middle of the pond walls to collect data on the growth status and group behavior of the farmed organisms. Monitoring units are installed on key equipment such as feeding equipment and water treatment equipment to collect operating parameters and status data in real time. Four temperature and humidity sensors, two light sensors, and one atmospheric pressure sensor are evenly deployed throughout the aquaculture workshop, covering the entire workshop area to ensure comprehensive environmental data collection. The edge layer also includes various devices for the execution module and edge computing nodes. These edge computing nodes utilize NVIDIA Jetson Xavier NX, with a processing latency of ≤100ms, enabling rapid response to local control needs. The edge layer employs a 5G communication module with a data transmission rate of ≥100Mbps, connecting the edge layer and the cloud layer. It is responsible for data preprocessing, command forwarding, and local collaborative control, avoiding the impact of data transmission latency on control effectiveness. The cloud layer deploys a big data management and control platform based on Hadoop architecture, with a storage capacity of ≥10TB, supporting distributed storage and parallel computing. This platform is used for data storage, model training, global control decision-making, and remote monitoring, meeting the processing and long-term storage needs of massive aquaculture data.

[0022] During system operation, multi-dimensional data is first collected through the data acquisition module. Water quality data is collected via deployed sensors, including dissolved oxygen, ammonia nitrogen, water temperature, pH, nitrate nitrogen, and CO2 concentration. The ammonia nitrogen sensor measures from 0 to 1 mg / L with an accuracy of ±0.01 mg / L; the water temperature sensor measures from 0 to 40℃ with an accuracy of ±0.2℃; and the pH sensor measures from 6.0 to 9.0 with an accuracy of ±0.1. These parameters are determined based on the suitable water quality range for the Litopenaeus vannamei, accurately reflecting the water quality status. Biological data is collected through machine vision equipment, acoustic sensors, and inspection robots. Machine vision equipment captures images of the feeding and swimming trajectories of the farmed organisms; acoustic sensors collect the sound wave signals of the organisms' activities; and the inspection robot patrols every 2 hours along a preset path, collecting data on population biomass and physiological indicators related to disease warnings, such as feeding intensity and swimming frequency. Equipment factor data is collected through monitoring units on each piece of equipment, including operating parameters and status data such as the feeding rate of the feeding equipment, the flow rate of the water treatment equipment, the operating location of the inspection equipment, and the cleanliness of the cleaning equipment. Environmental factor data is collected through sensors within the workshop, including workshop temperature and humidity, light intensity, and atmospheric pressure. The temperature and humidity sensors have a measurement range of 10-40℃ and 20%-90%RH, the light intensity sensor has a measurement range of 0-5000 lux, and the atmospheric pressure sensor has a measurement range of 80-120 kPa, ensuring comprehensive capture of changes in the workshop environment.

[0023] The collected multi-dimensional factor data is transmitted to the data processing and modeling module, where it undergoes preprocessing. A moving average method is used to smooth and denoise the data, eliminating noise caused by water body disturbance. The formula is as follows: ,in These are the values ​​after noise reduction. The original sampled values ​​are represented by n, which is the size of the moving window. Here, n is set to 5, a value determined based on the high-frequency characteristics of water body interference noise, effectively filtering noise without losing valid signals. Outlier removal is performed based on the 3σ statistical principle. The mean μ and standard deviation σ of the data are calculated. Data values ​​exceeding the range [μ-3σ, μ+3σ] are considered outliers, eliminating erroneous data caused by sensor drift. This method is a commonly used outlier identification method in statistics, adapted to the random error characteristics of sensor data. Linear interpolation is used to complete missing values ​​in the data. For cases where no more than three consecutive data points are missing, a linear function is constructed using two adjacent valid data points. , where a is the slope and b is the intercept, calculate the values ​​of missing data points to ensure that a complete standardized dataset is generated to meet the needs of subsequent modeling.

[0024] After data preprocessing, feature extraction is performed. Dynamic variation characteristics of water quality indicators are extracted from water quality factor data, such as the diurnal fluctuation range of dissolved oxygen, calculated as the difference between the maximum and minimum dissolved oxygen values ​​over 24 hours, reflecting water quality stability. Biological behavior characteristics are extracted from biological factor data, including feeding evenness and frequency of abnormal behavior. Feeding evenness is determined by analyzing the distribution of feeding areas of farmed animals captured by machine vision equipment, calculating the deviation between the proportion of feeding areas and the even distribution threshold. Abnormal behavior frequency is determined by matching the acoustic signals of farmed animals' activities collected by acoustic sensors with normal behavior acoustic templates, counting the number of mismatched signals. Equipment operation characteristics are extracted from equipment factor data, such as the correlation curve between aerator energy consumption and dissolved oxygen enhancement efficiency. By recording the energy consumption of aerators at different power levels and the corresponding dissolved oxygen enhancement, a correlation curve is fitted to provide a basis for subsequent energy consumption control.

[0025] After feature extraction, a multi-factor interaction model is constructed, along with three sub-models. The water quality dynamic prediction model uses a multiple linear regression model, combining factors such as feeding amount, stocking density, and water temperature, to predict the future trend of water quality indicators. The formula is as follows: ,in Let F be the predicted water quality index at time t, D be the stocking density, T be the water temperature, α, β, and γ be the weighting coefficients, and ε be the error term. The weighting coefficients are obtained by fitting historical aquaculture data using the least squares method; for example, α is set to 0.3, β to 0.4, and γ to 0.2. The threshold for ε is set to ±0.1, based on the priority analysis of the impact of feeding amount, stocking density, and water temperature on water quality in Litopenaeus vannamei farming. The growth prediction model uses a backpropagation (BP) neural network. The input layer consists of feed amount, water quality parameters (dissolved oxygen, ammonia nitrogen, pH), and aquaculture environment data (workshop temperature and humidity, light intensity). There are three hidden layers, each with 10 neurons. The output layer is the weight growth rate of the cultured organism. The activation function is the sigmoid function, the training iterations are 1000, and the learning rate is 0.01. These parameter settings are based on the conventional application experience of BP neural networks in biological growth prediction and are determined in conjunction with the sample size of the aquaculture data to ensure the prediction accuracy of the model. The biomass estimation model integrates the instance segmentation results from machine vision with the output of a density estimation network. Instance segmentation uses the Mask R-CNN algorithm to segment and identify individual cultured objects. The density estimation network uses a CNN network, taking the segmented image features as input and outputting the population density. The biomass estimation formula is as follows: Where B is the population biomass. Let be the average weight of the i-th cultured object. Let be the density of the i-th region, and n be the number of segmented regions. This model enables online estimation of population biomass. A multi-factor interaction model is constructed based on the outputs of three sub-models, analyzing the interaction patterns among water quality, organisms, equipment, and the environment, providing a core basis for regulatory decisions.

[0026] The multi-factor joint regulation decision module generates coordinated regulation instructions based on the output of the multi-factor interaction model and preset aquaculture targets. The preset aquaculture targets are determined according to the growth requirements of Litopenaeus vannamei, such as maintaining dissolved oxygen at 5-7 mg / L, ammonia nitrogen ≤0.2 mg / L, water temperature at 26-28℃, maintaining the weight gain rate of the cultured organisms within a reasonable range, and controlling energy consumption at a low level within the industry. The water quality regulation strategy dynamically adjusts the aerator power, water treatment system operating parameters, and water exchange flow rate based on the prediction results of the water quality dynamic prediction model and the biological requirements reflected by biological factor data. The formula for adjusting the aerator power is as follows: Where P is the adjusted power of the aerator. The current power is given, and k is the adjustment coefficient. For water quality target values, The value is a predicted value, with k set to 0.5. This is based on the linear relationship between aerator power and dissolved oxygen increase rate. When the predicted dissolved oxygen is lower than the target value, the aerator power is increased according to this formula to ensure that the dissolved oxygen quickly reaches the target. The feeding regulation strategy is based on the estimation results of the biomass estimation model, the prediction results of the growth prediction model, and the biological behavior characteristics. The feeding amount, feeding area, and feeding frequency are adjusted accordingly. The feeding amount is calculated as follows: Where r is the feed intake rate of the cultured organism and f is the feed conversion ratio. r is determined based on the weight growth rate of the growth prediction model. For example, when the weight growth rate is 5%, r is 0.08 and f is 0.7. The values ​​are based on the feeding characteristics of Litopenaeus vannamei at different growth stages. The feeding area is also adjusted according to the feeding uniformity. If the feed intake rate in a certain area is low, the feed amount in that area is reduced to avoid feed waste. The equipment coordination strategy uses a time-series planning algorithm to establish an equipment operation timeline, coordinating the operation sequence of feeding equipment, aeration equipment, and cleaning equipment. For example, the feeding equipment operates from 9:00 to 10:00, and the cleaning equipment operates from 10:30 to 11:30, avoiding conflicts caused by simultaneous operation of equipment in the same area and ensuring a smooth culture process. Energy consumption control strategies are based on real-time energy consumption data and aquaculture needs. They adjust the operating hours and power of high-energy-consuming equipment. For example, aerators, as high-energy-consuming equipment, can have their power appropriately increased during off-peak hours (00:00-06:00) when electricity prices are lower. This can reduce aquaculture costs while meeting dissolved oxygen requirements. The strategy is based on the peak-valley electricity pricing policy of the power grid and the diurnal variation of dissolved oxygen requirements during aquaculture.

[0027] Furthermore, the multi-factor joint regulation decision-making module also performs self-learning adjustments, acquiring historical regulation data and aquaculture effect feedback data provided by the feedback optimization module. It employs the DQN reinforcement learning algorithm, with the state space consisting of multi-dimensional factor data, the action space being a set of regulation instructions, and the reward function being... Where S is the aquaculture effect evaluation score, E is the energy consumption cost, and a and b are weighting coefficients, with a set of 0.7 and b set of 0.3, based on the need to balance aquaculture benefits and energy consumption costs. Based on historical control data and aquaculture effect feedback data, the parameters in the water quality dynamic prediction model, growth prediction model, biomass estimation model, and multi-factor interaction model are iteratively adjusted. The model parameters are updated every 10 control cycles, and the optimized model parameters are applied to the generation of the next round of collaborative control instructions, enabling the model to continuously adapt to changes in the aquaculture environment and improve control accuracy.

[0028] After receiving the coordinated control command, the execution module drives the corresponding aquaculture equipment to perform control actions. The water quality control execution unit controls the aerator, water pump, biofilter, disinfection equipment, and decarbonization device to perform dissolved oxygen enhancement, ammonia nitrogen removal, and water disinfection operations. The aerator is a jet aerator with a power range of 1.5-5.5kW. A 3kW aerator is selected for a 500m³ aquaculture pond, based on the matching relationship between water volume and aeration efficiency. The biofilter uses nitrifying bacteria filter media, with the filter media filling volume being 10% of the aquaculture pond's water volume to ensure that the ammonia nitrogen treatment efficiency meets the requirements. The feeding execution unit controls a track-mounted intelligent feeding robot to perform precise feeding and zoned feeding operations. The robot's movement speed is 0.5m / s, and the feeding accuracy is ±5g. During zoned feeding, the aquaculture pond is divided into 4 areas, with a feeding interval of 10 minutes between each area, based on the size of the aquaculture pond and the distribution of aquaculture species, to ensure uniform feeding. The equipment collaborative execution unit controls inspection robots, cleaning robots, and material handling robots, completing inspection, cleaning, and material handling tasks according to the timing sequence in the collaborative control instructions. The inspection robot's movement speed is 0.8 m / s, and the inspection cycle is 2 hours. The cleaning robot's cleaning rate is 20 m³ / h, adapting to the workshop layout and pollution generation rate of the aquaculture factory. The environmental control execution unit controls the workshop's air conditioning and lighting equipment, adjusting the workshop's temperature, humidity, and light intensity to the set range. The workshop air conditioning has a cooling capacity of 5 kW and a heating capacity of 3 kW. The lighting equipment uses LED lights, with a light intensity adjustment range of 500-2000 lux, meeting the temperature, humidity, and light conditions required for the growth of Litopenaeus vannamei.

[0029] The feedback optimization module monitors the execution effect of control actions in real time, collects multi-dimensional factor data after control, compares it with preset aquaculture targets, and generates feedback signals. First, it assesses the degree of conformity between the multi-dimensional factor data after control and the preset aquaculture targets. It generates effect evaluation indicators by calculating the water quality indicator compliance rate, biological growth status deviation, and energy consumption change rate. The water quality indicator compliance rate is calculated as follows: ,in The number of water quality indicators required to meet the standards, The total number of water quality indicators; the deviation of biological growth status is... ,in This represents the actual growth state value. The target value is [value]; the rate of change in energy consumption is [rate]. ,in For current energy consumption, This refers to the energy consumption of the previous cycle. When the performance evaluation indicators show that a certain factor deviates from the preset aquaculture target, such as the water quality compliance rate being below 90%, the cause of the deviation is analyzed. If it is due to insufficient dissolved oxygen, a correction instruction is generated to adjust the power and running time of the aerator; if it is due to insufficient feeding leading to deviations in growth status, the feeding strategy is adjusted. Simultaneously, the control process data, performance evaluation indicators, and correction instructions are archived to a cloud-based MySQL database. The data storage period is one year, based on the data requirements of the aquaculture cycle and model iteration, providing sufficient data support for model iteration.

[0030] In summary, this invention can deeply couple water quality, biological, equipment and environmental factors, solving the problem of the disconnect between the control behavior and the actual needs of the cultured organisms in the prior art. It makes the control strategy more in line with the growth needs of Litopenaeus vannamei, effectively maintains the long-term homeostasis of the culture process, and improves the survival rate and culture benefits of the cultured organisms.

[0031] Example 2

[0032] To enable those skilled in the art to fully understand and implement the present invention, the specific implementation principle of the present invention will be further explained below in conjunction with a specific application scenario.

[0033] This embodiment selects the scenario of factory farming of turbot to verify the practical application of the multi-factor joint regulation system, and focuses on illustrating the adaptability and operation effect of the system in different farming species scenarios.

[0034] First, the system parameters were adjusted based on the aquaculture characteristics of turbot. Turbot is a bottom-dwelling fish, with an optimal growth temperature of 14-18℃. It is sensitive to light and has strict requirements regarding nitrite concentration in the water. Therefore, in the data acquisition module, nitrite concentration monitoring was added to the water quality data, using a nitrite sensor with a measurement range of 0-0.2 mg / L and an accuracy of ±0.005 mg / L. The sensor was deployed in the same location as the ammonia nitrogen sensor. In the biological factors data, body surface condition monitoring was added. Machine vision equipment was used to magnify and photograph the turbot's body surface, capturing features such as scale integrity and skin color as supplementary indicators for disease early warning. In the environmental factors data, the accuracy of light intensity acquisition was improved, and the sensor measurement range was adjusted to 0-3000 lux to ensure accurate capture of changes in low-light environments.

[0035] The system deployment still adopts a three-tiered "cloud-edge-device" architecture. At the edge layer, the sensor deployment density is adjusted based on the layout of eight 600m³ turbot culture ponds. Each pond has three dissolved oxygen sensors and two nitrite sensors, evenly distributed at the bottom to meet the water quality monitoring needs of benthic fish. Machine vision equipment is deployed above the ponds, with a downward shooting angle for easy observation of the turbot's surface condition and feeding behavior. The hardware configuration of the edge and cloud layers is consistent with Example 1, except that the data transmission priority is adjusted based on the characteristics of the turbot's culture data, prioritizing biological surface condition data to ensure rapid transmission of disease early warning signals.

[0036] In the data processing and modeling module, the data preprocessing method remains unchanged, but the calculation range of the mean and standard deviation for outlier removal is adjusted for the newly added nitrite concentration data to ensure data validity. In the feature extraction stage, a new indicator of the proportion of abnormal body surface features is added from the biological factor data. Machine vision image analysis is used to statistically determine the proportion of turbot with abnormal body surface conditions. The sub-model parameters of the multi-factor interaction model are specifically adjusted. Nitrite concentration is added as an input variable to the water quality dynamic prediction model, and the weighting coefficients are refitted using historical turbot farming data; for example, the weighting coefficient for nitrite concentration is set to 0.25. The growth prediction model adds a nitrite concentration parameter to its input layer, and the number of neurons in the hidden layer is adjusted to 12 to adapt to the feature learning requirements of the newly added input variable. In the biomass estimation model, due to the flat body shape of turbot, the feature extraction algorithm for instance segmentation is optimized to improve the accuracy of individual identification.

[0037] The multi-factor joint control decision module adjusts its control strategies according to the aquaculture needs of turbot. In the water quality control strategy, the target value for nitrite concentration is set at ≤0.05 mg / L. When the predicted value exceeds the target range, the water exchange flow rate is increased, and the enhanced treatment mode of the biological filter is activated to accelerate nitrite removal. In the feeding control strategy, turbot are nocturnal feeders, so the feeding frequency is adjusted to twice a day, at 19:00 and 7:00 respectively. The feeding area is concentrated in the middle of the pond, and the feeding amount is adjusted based on the biomass estimation results and the output of the growth prediction model. The feeding rate r is set to 0.06 during nighttime feeding and 0.04 during the daytime to match their feeding patterns. In the equipment coordination strategy, the operating time of the cleaning robot is adjusted to 12:00-13:00 during the daytime to avoid disturbing the turbot's feeding and resting at night. In the energy consumption control strategy, since turbot thrive in lower water temperatures, the workshop air conditioning operates for longer periods during the summer. At the same time, the power grid's off-peak hours are used to store chilled water, thereby reducing energy costs.

[0038] The equipment parameters of the execution module were adjusted accordingly. The light intensity adjustment range of the lighting equipment was changed to 300-1500 lux, which is in line with the light-avoidance growth characteristics of turbot. The track-type intelligent feeding robot of the feeding execution unit reduced its movement speed to 0.3 m / s during nighttime feeding to improve feeding accuracy. In the feedback optimization module, the threshold for biological growth status deviation was adjusted to ±0.05, based on the fact that turbot's growth rate is relatively slow, and the deviation threshold needs to be adapted to its growth characteristics. A new body surface health compliance rate was added to the effect evaluation indicators. The percentage of turbot with normal body surface condition is counted. When this indicator is below 95%, a correction instruction related to disease prevention and control is generated to adjust the operating parameters of the disinfection equipment or the nutrient ratio in the feed.

[0039] During system operation, sensors in the edge layer collect real-time data on water quality, organisms, equipment, and the environment in the turbot farming ponds. After preprocessing by the edge layer, this data is transmitted to the cloud layer. The cloud layer's multi-factor interaction model analyzes the collected data, generates targeted collaborative control commands, and forwards them to the edge layer's execution module, driving various devices to perform control actions. The feedback optimization module continuously monitors the control effect, collects various data after control, compares them with preset farming targets, analyzes the causes of deviations, generates correction commands, and archives relevant data for model iteration.

[0040] Through application verification in the industrialized farming scenario of turbot, the system can quickly adapt to the characteristic requirements of different farmed species, achieve precise joint regulation of multiple factors, effectively maintain the stability of the farming environment, and provide a guarantee for the healthy growth of farmed organisms, demonstrating the system's versatility and practicality.

[0041] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-factor joint regulation system based on aquaculture farms, characterized in that, include: The data acquisition module is used to collect multi-dimensional factor data that affect the aquaculture homeostasis within the aquaculture factory. The multi-dimensional factor data includes at least water quality factor data, biological factor data, equipment factor data, and environmental factor data. The data processing and modeling module is used to receive the multi-dimensional factor data, perform data preprocessing and feature extraction, and construct a multi-factor interaction model. The multi-factor interaction model is used to analyze the interaction patterns among water quality, organisms, equipment and environment. The multi-factor joint regulation and decision-making module is used to generate coordinated regulation instructions based on the output results of the multi-factor interaction model and the preset breeding target. An execution module is used to receive the coordinated control instructions and drive the corresponding aquaculture equipment to perform control actions; The feedback optimization module is used to monitor the execution effect of the control action, collect multi-dimensional factor data after control, and compare the multi-dimensional factor data after control with the preset breeding target to form a feedback signal for correcting the multi-factor interaction model or collaborative control command.

2. The multi-factor joint control system based on aquaculture farms according to claim 1, characterized in that, The data acquisition module is specifically used for: The water quality data, including dissolved oxygen, ammonia nitrogen, water temperature, pH value, nitrate nitrogen, and CO2 concentration, are collected by sensors deployed in aquaculture ponds and water treatment systems. The biological factor data is collected using machine vision equipment, acoustic sensors, and inspection robots. The biological factor data includes the growth status of the cultured objects, group behavior, group biomass, and physiological indicators related to disease early warning. The equipment factor data is collected through the monitoring units on the feeding equipment, water treatment equipment, inspection equipment and cleaning equipment. The equipment factor data includes the operating parameters and status data of each piece of equipment. The environmental data, including workshop temperature and humidity, light intensity, and atmospheric pressure, are collected by sensors deployed in the breeding workshop.

3. The multi-factor joint control system based on aquaculture farms according to claim 2, characterized in that, The data processing and modeling module, during data preprocessing, is specifically used for: The multi-dimensional factor data is subjected to smoothing and denoising processing to eliminate noise caused by water body interference; Outlier removal was performed on the multi-dimensional factor data based on statistical principles to eliminate erroneous data caused by sensor drift. Interpolation is used to fill in the missing values ​​in the multi-dimensional factor data to generate a complete standardized dataset.

4. The multi-factor joint control system based on aquaculture farms according to claim 3, characterized in that, The data processing and modeling module, when performing feature extraction, is specifically used for: The dynamic change characteristics of water quality indicators are extracted from the water quality factor data, including the diurnal fluctuation range of dissolved oxygen. Biological behavioral characteristics are extracted from the biological factor data, including feeding evenness and frequency of abnormal behavior; The equipment operating characteristics are extracted from the equipment factor data, including the correlation curve between the oxygenator's energy consumption and dissolved oxygen enhancement efficiency.

5. The multi-factor joint control system based on aquaculture farms according to claim 4, characterized in that, The data processing and modeling module is also used to construct the following sub-models when building a multi-factor interaction model: A dynamic water quality prediction model is used to predict future trends in water quality indicators by combining factors such as feeding amount, stocking density, and water temperature. Growth prediction models are used to predict the future growth status of farmed organisms by combining feed input, water quality parameters, and aquaculture environment data. A biomass estimation model is used to fuse instance segmentation results from machine vision with the output of a density estimation network to estimate population biomass online. The multi-factor interaction model is constructed based on the outputs of the water quality dynamic prediction model, the growth prediction model, and the biomass estimation model.

6. The multi-factor joint control system based on aquaculture farms according to claim 5, characterized in that, When generating coordinated control instructions, the multi-factor joint regulation decision-making module is specifically used to generate the following strategies: A water quality control strategy is used to dynamically adjust the aerator power, water treatment system operating parameters, and water exchange flow rate based on the prediction results of the water quality dynamic prediction model and the biological needs reflected by the biological factor data. Feeding regulation strategy, used to adjust feeding amount, feeding area and feeding frequency based on the estimation results of the biomass estimation model, the prediction results of the growth prediction model and the biological behavior characteristics; Equipment coordination strategy is used to coordinate the operation sequence of feeding equipment, oxygenation equipment and cleaning equipment to achieve conflict-free operation of equipment; Energy consumption control strategies are used to adjust the operating time and power of high-energy-consuming equipment based on real-time energy consumption data and aquaculture needs.

7. The multi-factor joint control system based on aquaculture farms according to claim 6, characterized in that, The multi-factor joint regulation and decision-making module is also used to perform self-learning adjustment, specifically in the following manner: Acquire historical control data and aquaculture effect feedback data provided by the feedback optimization module; Using reinforcement learning algorithms, the parameters in the water quality dynamic prediction model, growth prediction model, biomass estimation model, and multi-factor interaction model are iteratively adjusted based on the historical regulation data and aquaculture effect feedback data. The optimized model parameters will be applied to the generation of the next round of coordinated control commands.

8. The multi-factor joint control system based on aquaculture farms according to claim 1, characterized in that, The execution module includes: The water quality control execution unit is used to control the aerator, water pump, biological filter, disinfection equipment and decarbonization device to perform dissolved oxygen enhancement, ammonia nitrogen removal and water disinfection operations; The feeding execution unit is used to control the track-mounted intelligent feeding robot to perform precise feeding and zoned feeding operations; The equipment collaborative execution unit is used to control the inspection robot, the cleaning robot, and the handling robot to complete the inspection, cleaning, and material handling tasks according to the timing sequence in the collaborative control instructions. The environmental control execution unit is used to control the workshop's air conditioning and lighting equipment, adjusting the workshop's temperature, humidity, and light intensity to the set range.

9. The multi-factor joint control system based on aquaculture farms according to claim 1, characterized in that, When generating a feedback signal, the feedback optimization module is specifically used for: The assessment evaluates the alignment between multi-dimensional data after regulation and the preset aquaculture targets. By calculating the compliance rate of water quality indicators, the deviation of biological growth status, and the change rate of energy consumption, an effect evaluation index is generated. When the effect evaluation index shows that a certain factor index deviates from the preset breeding target, the cause of the deviation is analyzed and a correction instruction for adjusting the relevant control strategy is generated. Data on the regulation process, effect evaluation indicators, and correction instructions are archived in a cloud database to provide data support for model iteration.

10. The multi-factor joint control system based on aquaculture farms according to claim 1, characterized in that, The system is deployed using a three-tier architecture: cloud-device-edge. The edge layer, deployed at the breeding farm site, includes the field acquisition equipment, execution module, and edge computing nodes in the data acquisition module, and is used to perform real-time data acquisition and local rapid regulation; The cloud layer is equipped with a big data management and control platform, which is used to perform data storage, model training, global control decision-making, and remote monitoring. The edge layer connects the edge layer and the cloud layer, and is used to perform data preprocessing, instruction forwarding, and local collaborative control.

Images