Rice and fish integrated planting and breeding system and method based on multi-source data fusion and intelligent collaborative decision

Abstract

The invention discloses a rice and fish integrated planting and breeding system and method based on multi-source data fusion and intelligent collaborative decision, and the system comprises an engineering rice field facility, a multi-source sensing network, an intelligent execution mechanism, an edge control unit and a cloud decision platform, the multi-source sensing network is used for collecting water quality, soil, weather and biological vision multi-source heterogeneous data; the intelligent execution mechanism executes the production action of rice and fish planting; the edge control unit is connected with the multi-source sensing network and the intelligent execution mechanism, realizes data aggregation, instruction analysis and real-time control, and interacts with the cloud decision platform through a remote communication network; the cloud decision-making platform deploys rice-fish multi-source data fusion to generate a decision-making model, and generates a collaborative regulation and control strategy considering the rice-fish symbiotic synergistic effect and the long-term production benefit by using the model and the sensing data. The system provides a systematic solution for precise and intelligent management of rice and fish integrated planting and breeding.

Description

Rice-Fish Integrated Farming System and Method Based on Multi-Source Data Fusion and Intelligent Collaborative Decision-Making Technical Field

[0001] This invention belongs to the interdisciplinary field of smart agriculture and ecological aquaculture, specifically involving the integrated application of the Internet of Things, sensor technology, big data analysis, and artificial intelligence models in ecological agriculture, and particularly involving a rice-fish integrated farming system and method based on multi-source data fusion and intelligent collaborative decision-making. Background Technology

[0002] With the advancement of agricultural modernization, the Internet of Things (IoT) and automation technologies have been initially applied in integrated rice-fish farming, forming a series of technical solutions centered on environmental monitoring and automatic equipment control. Currently, existing technologies in this field mainly focus on integrated monitoring system design. Related patent documents disclose the construction of a remote monitoring platform based on wireless transmission by deploying water quality monitoring equipment such as dissolved oxygen sensors, water temperature sensors, and pH sensors, and linking them with actuators such as aerators and feeders. Such systems can achieve real-time data acquisition and visualization of key water quality parameters, and trigger corresponding control commands based on preset fixed thresholds, thereby shifting the traditional production model relying on manual pond inspections and experience-based judgment to an automated response model based on data. Other technical solutions focus on optimizing aquaculture engineering facilities, such as designing dedicated circulating water ditches or purification structures, aiming to improve the water's self-purification capacity and stocking density. These existing technologies collectively represent the mainstream trend of rice-fish farming towards digitalization and engineering, providing a preliminary technical path to solve the problem of extensive management.

[0003] Despite the progress made in existing technologies, significant limitations remain in their inherent technical logic and architecture, hindering further improvements in management efficiency and system effectiveness.

[0004] First, at the data perception level, existing technical solutions suffer from limitations in dimensionality and information fragmentation. Most systems focus only on monitoring direct parameters of the aquaculture water body, lacking simultaneous and systematic collection of key indicators of paddy field soil (such as nitrogen, phosphorus, and potassium nutrient content, moisture status, and pH value) and field microclimate elements (such as light intensity, wind speed, rainfall, and evapotranspiration). This results in decision-making based on fragmented aquatic environmental data, failing to comprehensively reflect the growth needs of rice, soil fertility capacity, and climate-driven influences, leading to a one-sided decision-making basis.

[0005] Secondly, at the control logic level, existing technologies generally exhibit discrete, single-factor response mechanisms. Irrigation, feeding, aeration, and temperature regulation are all independent control loops, following simple "condition-action" rules. This mechanism lacks coordination strategies between subsystems, easily leading to management conflicts. For example, when adjusting field water levels to meet the water management requirements of rice at specific growth stages, failure to simultaneously consider and automatically adjust fish stocking density, feeding strategies, or water aeration programs may increase the risk of space stress for fish or water quality deterioration.

[0006] Finally, at the decision-making mechanism level, existing technologies are essentially static and passive threshold-response modes. All control thresholds must be set manually in advance based on experience; the system cannot adaptively adjust according to the dynamic changes in the growth stages of rice, the physiological needs of fish at different growth stages, or short-term weather trends. Its operation is limited to post-event remediation of abnormal states and lacks the ability to utilize multi-source historical and real-time data for modeling analysis, trend prediction, risk assessment, and multi-objective optimization decision-making. Therefore, existing technologies struggle to achieve the leap from automation to intelligence, exhibiting significant technical bottlenecks in improving resource utilization efficiency, mitigating potential production risks, and maximizing overall system benefits. Summary of the Invention

[0007] This invention aims to address the technical shortcomings of existing rice-fish integrated farming systems, as pointed out in the background art, such as single data perception dimension, discrete control logic, and static and passive decision-making mechanism, by providing a rice-fish integrated farming system and method based on multi-source data fusion and intelligent collaborative decision-making.

[0008] The objective of this invention is achieved through the following technical solution: a rice-fish integrated farming system based on multi-source data fusion and intelligent collaborative decision-making, comprising: engineered paddy field facilities, a multi-source sensing network, intelligent actuators, an edge control unit, and a cloud-based decision-making platform. The engineered paddy field facilities include aquaculture ditches, controllable inlets and outlets, a paddy field planting area, and controllable interconnecting inlets between the aquaculture ditches and the paddy field planting area. The multi-source sensing network is deployed in and around the paddy field to simultaneously collect heterogeneous multi-source data on water quality, soil, meteorology, and biological vision. The intelligent actuators include oxygenation, heating, feeding, and water circulation control equipment, executing rice-fish farming production actions according to control commands issued by the edge control unit. The edge control unit has built-in programmable... The programmable logic controller (PLC) connects to a multi-source sensing network and an intelligent actuator via a local communication network to achieve data aggregation, command parsing, and real-time control. It also interacts with a cloud-based decision-making platform via a remote communication network. The cloud-based decision-making platform deploys a rice-fish multi-source data fusion generation decision model. This model incorporates a rice full-growth-cycle knowledge base and a fish growth-environment response model library. Utilizing the rice-fish multi-source data fusion generation decision model and sensing data collected by the multi-dimensional sensing network, it generates a collaborative regulation strategy that balances the synergistic effect of rice-fish symbiosis with long-term production benefits. The collaborative regulation strategy is then distributed to the edge control unit, enabling the edge control unit to drive the intelligent actuator to execute the collaborative regulation strategy. The platform also receives feedback data after strategy execution to complete autonomous model optimization.

[0009] Furthermore, the aquaculture ditch surrounds the paddy field planting area and has multiple controllable interconnecting water inlets, which are spaced apart along the side wall of the aquaculture ditch. Each interconnecting water inlet is equipped with a flat gate that can be raised and lowered by a motor, and the opening degree and opening time of the flat gate are controllable.

[0010] Furthermore, the multi-source sensing network includes a water quality monitoring module, a soil monitoring module, a meteorological monitoring module, and a visual monitoring module. The water quality monitoring module is deployed at points with stable water flow in the aquaculture ditches to monitor dissolved oxygen concentration, pH value, conductivity, water temperature, and water level changes within the ditches. The soil monitoring module is deployed in the paddy field planting area to monitor soil volumetric water content, temperature, pH value, conductivity, and nutrient content. The meteorological monitoring module monitors air temperature, relative humidity, wind speed, wind direction, light intensity, rainfall, and atmospheric pressure in the fields. The visual monitoring module acquires visual data from the aquaculture ditches and paddy fields.

[0011] Furthermore, the intelligent actuator includes an ultrafine bubble aeration pipe network and a heating pipe network laid at the bottom of the aquaculture ditch, an automatic feeder installed at the edge of the aquaculture ditch, and a water circulation control system consisting of an irrigation pump, a circulation pump, and an electric inlet and outlet valve.

[0012] This invention also provides a method for integrated rice-fish farming based on multi-source data fusion and intelligent collaborative decision-making. Using the system described above, the method includes the following steps: S1: Synchronous acquisition of multi-source heterogeneous data. Multi-source heterogeneous data on water quality, soil, meteorology, and biological vision are synchronously acquired through a multi-source sensing network. This data is uploaded to an edge control unit and then synchronized to a cloud-based decision-making platform. S2: Multi-source data fusion and digital twin construction. The cloud-based decision-making platform preprocesses the received multi-source heterogeneous raw data and extracts key features reflecting the system state, generating a spatiotemporally consistent system panoramic feature dataset and constructing a dynamic digital twin of the rice-fish symbiotic system. S3: Knowledge-based dynamic supply and demand diagnosis. The cloud-based decision-making platform calls the built-in rice full-growth-cycle knowledge base and fish growth-environment response model library, combining the real-time system state represented by the dynamic digital twin, to update the dynamic target relationship matrix and quantify the mutual influence between multiple management targets. S4: Co-effect modeling and collaborative strategy planning. Based on the updated dynamic target relationship matrix, a co-effect function incorporating the synergistic effect between targets is constructed, and a state transition mechanism is used to perform the modeling. The sequential decision engine simulates the impact of feasible control action combinations on the system state within a future preset decision time domain, calculates the long-term expected synergistic effect value of each action sequence, selects the optimal action sequence, and generates a collaborative regulation strategy package; S5: Strategy distribution and precise execution. The collaborative regulation strategy package is distributed to the edge control unit via the cloud decision platform. After the programmable logic controller built into the edge control unit parses the instructions, it drives the intelligent actuator to complete operations such as irrigation, feeding, oxygenation, temperature regulation, and fish spatial scheduling; S6: Safety closed-loop verification. During the strategy execution process, if high-risk agricultural operations such as fertilization or pesticide application are identified, a safety isolation process is automatically triggered. Key harmful indicators in the field water are monitored at high frequency through a sensing network. Trend analysis is performed in conjunction with a fish toxicology safety threshold library. The isolation is lifted after verifying that the environment is safe; otherwise, the isolation is maintained and an alarm is triggered; S7: Model evolution learning. A complete data closed loop of "real-time system state - collaborative regulation strategy - new state after execution - actual production utility" is collected. Based on this data closed loop, incremental training and parameter optimization are performed on the rice-fish multi-source data fusion growth decision model of the cloud decision platform to achieve autonomous model evolution.

[0013] Furthermore, the preprocessing includes noise filtering, missing value interpolation, timestamp alignment, and protocol unification.

[0014] Furthermore, the key features extracted included soil nitrogen content, dissolved oxygen concentration in water, NDVI index of rice canopy, and fish activity intensity index.

[0015] Furthermore, the dynamic target relationship matrix is ​​as follows: ,in, Indicates at time Management Objectives Management objectives The impact coefficient is determined by multiple management objectives, including high rice yield, fish health, resource conservation, and ecological balance.

[0016] Furthermore, the shared utility function is: ,in, Let X represent the total number of management objectives, and let X represent different control strategies. Indicates the first time when selecting control strategy X. The single-objective utility of a management objective Indicates the first The weight of each management objective, Represents the synergy coefficient. The function is used to measure the joint utility between two pairs of objectives.

[0017] Furthermore, S4 specifically includes: S41: Obtaining the current state of the paddy field ecosystem. That is, a dynamic digital twin, which represents the current state. Input a sequential decision engine with state transition functionality to formalize the future rice-fish farming management problem into a Markov decision process with a finite time domain; S42: Define the core elements of the Markov decision process, including the preset decision time domain, state space, and action space. and state transition function, wherein the decision time domain is set to 24 hours, and the action space The state transition function, defined by the system's built-in system dynamic model, describes the execution actions and includes all adjustable variables and their value ranges, such as irrigation amount, feeding amount, aeration intensity, and gate opening. After state from Transfer to The probability and outcome; S43: in the current state As the root node of the strategy tree, from the action space Extract all possible actions in the current state, and use each possible action as a branch edge of the root node. Simulate the next state after each action is executed through the state transition function. S44: Split the decision time domain according to the preset decision step size, recursively expand the policy tree, and for nodes that have not been expanded in the policy tree, extract the possible actions in the corresponding state and simulate the state transition to generate child nodes in turn, until the expansion depth of all nodes covers the entire decision time domain, forming a policy tree containing all possible action sequences. Each path from the root node to the leaf node corresponds to a set of continuous control action sequences in the decision time domain; S45: Use a simulation-based search algorithm to traverse and evaluate the paths in the policy tree through four iterations: selection, expansion, simulation, and backtracking, and calculate the long-term expected synergistic effect of each path. ,in, , This represents the total number of decision steps broken down within the decision-making time domain. Indicates the discount factor. Indicates the first The prediction co-effect in the step state is achieved by Calculated; S46: Selected The path with the highest value is the optimal coordinated control action sequence within the decision-making time domain; S47: The optimal coordinated control action sequence is packaged into a coordinated control strategy package through a strategy encapsulation algorithm. The coordinated control strategy package specifies the timing operation parameters of each intelligent actuator. The operation parameters include: execution time, control intensity, and runtime. The coordinated control strategy package also includes timing and temperature control instructions to guide fish to migrate directionally between aquaculture ditches and paddy fields, ensuring that the actions of each intelligent actuator are coordinated without conflict and meet the needs of rice-fish symbiosis.

[0018] The beneficial effects of this invention are as follows: 1. In terms of core decision-making capabilities, this invention achieves a fundamental leap from passive response relying on fixed thresholds and human experience to proactive prediction and optimization based on panoramic data perception and model calculation; by deeply integrating water quality, soil, meteorological, and biological visual information, a dynamic digital twin of the paddy field ecosystem is constructed. Its intelligent diagnosis and multi-objective optimization mechanism can coordinate the previously isolated production links such as irrigation, feeding, environmental control, and fish ecological scheduling, generating a strategy package that is highly coordinated in time and logic. This completely solves the technical problems of fragmented control logic and mutual constraints between subsystems in traditional automated systems, marking a shift in management paradigm from automation to intelligence. 1. Evolution of the system; 2. In terms of resource utilization and production efficiency, this solution achieves efficient utilization of water resources, feed, fertilizer and energy through precise perception and system optimization; Model-driven variable operations can supply water and fertilizer according to the water and fertilizer requirements of rice at different growth stages and the real-time growth needs of fish, significantly reducing the waste of resource input; At the same time, the system's predictive regulation of key environmental factors such as dissolved oxygen and temperature in the water effectively avoids the occurrence of stress conditions in the production environment, providing a continuous and stable optimal growth environment for plants and animals. This not only helps to ensure and improve the final yield and product quality of rice and fish, but also greatly enhances the overall stability and risk resistance of the agricultural production system.

[0019] 3. In terms of technology promotion and industrial upgrading, this invention successfully transforms the management knowledge in traditional agriculture, which relies heavily on personal experience and is difficult to articulate and replicate, into coded, calculable, and iteratively optimized digital models and software / hardware system standards. This transformation provides a practical technical path for the standardization and large-scale promotion of rice-fish integrated farming. The system significantly reduces the labor intensity of daily management and the reliance on specific skills, while improving the certainty and replicability of the production process, thus playing a positive role in promoting the modernization and industrial upgrading of ecological agriculture.

[0020] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings, in which: Figure 1 is a system overall architecture diagram of this invention; Figure 2 is a layout concept diagram of the system in this invention; Figure 3 is a hardware structure framework diagram of this invention; Figure 4 is a hardware structure block diagram of the water quality monitoring module; Figure 5 is a hardware structure block diagram of the soil monitoring module; Figure 6 is a hardware structure block diagram of the meteorological monitoring module; Figure 7 is a hardware structure block diagram of the water pump motor control module; Figure 8 is a schematic flowchart of the method of this invention; Figure 9 is a schematic flowchart of collaborative strategy planning; Figure 10 is a flowchart of intelligent water level regulation; and Figure 11 is a flowchart of water quality management and control. Detailed Implementation

[0022] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the preferred embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0023] This invention proposes a rice-fish integrated farming system and method based on multi-source data fusion and intelligent collaborative decision-making. This technical solution is based on engineered facilities, uses a "cloud-edge-device" collaborative architecture as its carrier, and employs a core algorithm model as its core, constructing a complete closed-loop system from physical perception to intelligent decision-making. This system is a deeply integrated hardware and software entity. Specifically, referring to Figures 1-3, a rice-fish integrated farming system based on multi-source data fusion and intelligent collaborative decision-making includes: engineered paddy field facilities, a multi-source sensing network, intelligent actuators, an edge control unit, and a cloud-based decision-making platform.

[0024] Figure 2 is an exemplary layout concept diagram. The engineered paddy field facility is an engineered paddy field, including aquaculture ditches (such as the ring-shaped aquaculture ditch in Figure 2), controllable inlets and outlets, a paddy field planting area, and controllable interconnecting inlets between the aquaculture ditch and the paddy field planting area. The aquaculture ditch surrounds the paddy field planting area, forming a ring structure. As an example, the aquaculture ditch can be 0.8 to 1.2 meters deep and 2 to 3 meters wide, forming the main habitat for fish. The ditch system is equipped with controllable inlets and outlets, each equipped with an electric inlet valve and outlet valve, and connected to an external water source and recycling treatment facility, providing infrastructure for precise water intake and replenishment. To achieve deep rice-fish symbiosis through refined management of fish activity, multiple controllable interconnecting inlets are set along the sidewall of the ditch (e.g., every 15-20 meters) between the ditch and the paddy field planting area. Each interchange is equipped with a flat gate that can be raised and lowered by a motor. Its opening degree and opening time can be precisely controlled by the control system, which is a key channel facility for fish to migrate directionally between their habitat and the work area.

[0025] At the perception level, the system also deploys a fully covered environmental monitoring network (i.e., a multi-source sensing network). The multi-source sensing network is deployed in paddy fields and surrounding areas to simultaneously collect multi-source heterogeneous data on water quality, soil, meteorology, and biological vision.

[0026] In some embodiments, the multi-source sensing network includes a water quality monitoring module, a soil monitoring module, a meteorological monitoring module, and a visual monitoring module.

[0027] The water quality monitoring module is deployed at points in the aquaculture ditch where water flow is stable (e.g., in the middle of the ditch, away from the inlet and outlet). It is used to continuously and in real-time monitor the dissolved oxygen concentration, pH value, conductivity, water temperature, and water level changes in the ditch. Figure 4 is a schematic diagram of the water quality monitoring module. As shown in Figure 4, this module may include a temperature sensor (to collect water temperature), a turbidity sensor (to collect water turbidity), a pH sensor (to collect water acidity / alkalinity data, such as pH value), a conductivity sensor (to collect water conductivity data, which reflects the water ion concentration), a dissolved oxygen sensor (to collect dissolved oxygen content data), and a water level sensor (to collect water level data). These sensors are all connected to the main control module (MCU) of the water quality monitoring module, transmitting the collected environmental parameters to the main control module in real time. The main control module, as the core computing and control unit, undertakes functions such as data reception, processing, and command forwarding, and serves as the communication hub between the sensors and external modules. The water quality monitoring module also includes a LoRa module, which connects to the main control module and serves as a remote communication interface to enable data transmission and exchange between the main control module and external systems (such as cloud platforms and edge control units). A battery (as a power supply module) stores solar energy to provide a stable power supply to the main control module and various sensors, ensuring continuous operation of the terminal. This water and soil monitoring module, through the integration of multi-dimensional sensors, a built-in main control module, and a remote communication interface, achieves real-time acquisition and remote transmission of water environment parameters (humidity, pH, conductivity, etc.) and water level parameters in rice-fish farming systems. It is a dedicated terminal device for water quality sensing in intelligent rice-fish farming systems.

[0028] The soil monitoring module is deployed in paddy field planting areas using a grid-based method, embedding integrated soil sensing nodes in the main root activity layer. These nodes continuously monitor soil volumetric water content, temperature, pH value, and electrical conductivity. Information on key nutrients such as nitrogen, phosphorus, and potassium in the soil is monitored in situ using dedicated ion-selective electrode sensors integrated into the nodes, or through periodic manual sampling for laboratory analysis. The data is then manually entered into a cloud management platform for supplementation and calibration. Figure 5 is a schematic diagram of the soil monitoring module. As shown in Figure 5, the module may include a conductivity sensor (collecting soil conductivity data, reflecting soil salinity and ion concentration), a temperature sensor (collecting soil temperature data), a moisture sensor (collecting soil moisture content data), a pH sensor (collecting soil acidity / alkalinity data, e.g., pH value), a nitrogen, phosphorus, and potassium sensor (collecting data on the content of nitrogen, phosphorus, and potassium nutrients in the soil), and a water level sensor (collecting water level data in related soil areas, such as field surfaces and ditches). All these sensors are connected to the main control module (MCU) of the soil monitoring module, transmitting the collected environmental parameters to the main control module in real time. The soil monitoring module also includes a LoRa module, which connects to the main control module and serves as a remote communication interface to enable data transmission and exchange between the main control module and external systems (such as cloud platforms and edge control units). A battery (as a power supply module) stores solar energy to provide a stable power supply to the main control module and various sensors, ensuring continuous operation of the terminal. Through the integration of multi-dimensional sensors, a built-in main control unit, a remote communication interface, and an independent power supply system, this soil monitoring module achieves real-time acquisition and remote transmission of soil environmental parameters (nutrients, temperature, humidity, pH, etc.) and related water level parameters in rice-fish farming systems. It is a dedicated terminal device for soil environmental sensing in intelligent rice-fish farming systems.

[0029] The meteorological monitoring module can collect data through localized meteorological data acquisition. For example, a meteorological monitoring module (such as a small automatic weather station) can be set up in open, unobstructed locations such as field ridges. It integrates multiple sensors to monitor air temperature, relative humidity, wind speed, wind direction, light intensity, rainfall, and atmospheric pressure in the field, providing a comprehensive understanding of the field's microclimate. Figure 6 is a schematic diagram of the meteorological monitoring module. As shown in Figure 6, the module may include a temperature sensor (collecting field air temperature data), a humidity sensor (collecting field air relative humidity data), a PM2.5 sensor (collecting field air PM2.5 concentration data), a wind speed sensor (collecting field air flow speed data), and a pressure sensor (collecting field atmospheric pressure data). All of these sensors are connected to the main control module (MCU) of the meteorological monitoring module, transmitting the real-time collected meteorological parameters to the main control module for processing. In some embodiments, the meteorological monitoring module can also receive data from the meteorological bureau via the Internet. The meteorological monitoring module also includes a LoRa module, which is connected to the main control module and serves as a remote communication interface to enable data transmission and exchange between the main control module and external systems (such as cloud platforms and edge control units). The battery (as a power supply module) can store solar energy to provide a stable power supply to the main control module and various sensors, ensuring continuous operation of the terminal.

[0030] Visual monitoring modules (e.g., 360° panoramic high-definition network cameras) can be installed on utility poles or dedicated poles in the fields. Their installation height and angle must ensure that the monitoring field of view effectively covers the main water surface areas of ditches and large areas of rice paddy canopy, used to acquire visual data of aquaculture ditches and rice paddies. This module processes the video stream in real time using image analysis models (such as image recognition algorithms based on convolutional neural networks), and can output quantitative information such as fish activity intensity index, fish distribution, and rice canopy color and texture features (such as the NDVI index), providing visual evidence for the analysis of biological behavior and environmental feedback.

[0031] At the execution level, the system is equipped with a series of intelligent devices, namely intelligent actuators. Referring to Figure 2, the intelligent actuators include oxygenation, heating, feeding, and water circulation control equipment, which execute the production actions of rice-fish farming according to the control commands issued by the edge control unit. The intelligent actuators include an ultra-fine bubble oxygenation pipe network and a heating pipe network (connected to a heat pump or boiler) laid at the bottom of the aquaculture ditch, an automatic feeder installed at the edge of the aquaculture ditch (i.e., the "automatic feeder" in Figure 2), and a water circulation control system consisting of an irrigation pump, a circulation pump, and electric inlet and outlet valves. For example, the aeration system includes: the main unit of the microbubble aerator is located in the field power distribution room or a dedicated equipment box; the main air supply pipeline is laid along the field ridges and connected to nano-aeration coils pre-laid evenly at the bottom of the aquaculture ditch via branch pipes; the heating system uses flexible, corrosion-resistant PE-RT pipes as heat exchange tubes, laid in parallel coils and fixed to the bottom and lower half of the side walls of the aquaculture ditch, and the pipelines converge and connect to an air source heat pump unit or electric heating boiler located on the bank; the automatic feeder is fixedly installed on a pre-cast concrete base at the edge of the aquaculture ditch, and its throwing outlet height and angle are adjusted to ensure that the feed can be evenly scattered on the surface of the target water area; a circulation pump is installed in the ditch to promote internal water exchange; an irrigation pump is installed at the water source; and all electric valves (inlet valves, outlet valves) are installed at the corresponding pipe interfaces. All of these devices are driven by a high-voltage electrical control circuit.

[0032] Figure 7 is a schematic diagram of the water pump control module. As shown in Figure 7, this module includes relays, a motor, and a main control module. The relays are connected to the main control module, receive control signals from the main control module, and realize the switching of the circuit; they are the "switching execution components" for the main control module to control the motor. The motor, as the power unit of the water pump, obtains AC power and receives on / off signals through the relays to realize the start and stop control of the water pump. This module also includes a LORA module, which is connected to the main control module of the water pump control module and serves as a remote communication interface to realize the command interaction between the main control module and external systems (such as cloud platforms, edge control units) and receive remote control commands. The battery (as a power supply module) provides operating power to the main control module and relays, ensuring the stable operation of the module.

[0033] All intelligent actuators are connected to an edge intelligent control cabinet located in the field. The edge control unit within this cabinet integrates an industrial programmable logic controller (PLC) as the local control core. It connects to a multi-source sensing network and intelligent actuators via a local communication network, enabling data aggregation, command parsing, and real-time control. It also interacts with a cloud-based decision-making platform via a remote communication network. Specifically, the edge control unit employs a dual-mode communication strategy: locally, it connects to various terminal devices (i.e., the sensing network and intelligent actuators) via a LoRA wireless self-organizing network or an RS-485 bus, achieving low-power, high-reliability access; remotely, it connects the edge control cabinet to the cloud-based decision-making platform via a 4G / 5G network (using the MQTT over TLS protocol), ensuring high-speed and stable command and data exchange. In the event of a network outage, the edge control unit possesses local caching and basic rule execution capabilities to ensure basic system operation.

[0034] In some embodiments of the present invention, the main control module in the multi-source sensing network and intelligent actuator of the system can use an STM32F1 microcontroller as the core computing unit, while the edge control unit can use an STM32H7 microcontroller as the core computing unit.

[0035] A multi-source data fusion model for rice-fish farming is deployed in the cloud-based decision-making platform. This model incorporates a knowledge base of the entire rice growth cycle and a fish growth-environment response model library. Utilizing the multi-source data fusion model and sensory data collected by a multi-dimensional sensing network, a collaborative regulation strategy is generated that balances the synergistic effect of rice-fish symbiosis with long-term production benefits. This collaborative regulation strategy is then distributed to the edge control unit, enabling the edge control unit to drive the intelligent actuator to execute the collaborative regulation strategy. Additionally, the model receives feedback data after the strategy is executed to complete autonomous optimization.

[0036] As shown in Figure 3, the system may also include a reserved wireless interface to support future functional expansion and upgrades. For example, an internet access interface can be provided to allow users to connect to the network for control and integrate with internet operations. The system may also include a power supply module. For example, a battery can provide power to the system, ensuring continuous operation. This battery can be charged using solar energy, achieving green functionality. Users can also view environmental data, issue control commands, and meet mobile management needs through terminals (e.g., web or mobile app).

[0037] Figure 2 illustrates the monitoring, control, and management facilities throughout the entire process. For example, a water flow meter is installed on the main inlet pipe to measure the total amount of water entering the rice-fish integrated farming base, enabling precise water resource measurement and cost accounting; a camera device is deployed above the farming base to monitor the growth status of rice and fish in real time, providing a visual management basis; a weather station is set up around the base to collect meteorological data such as field temperature, humidity, and wind speed, providing meteorological support for environmental control; inlet / outlet valves control the base's water intake and drainage respectively, working in conjunction with water level control logic to achieve precise water level management; a water-fertilizer mixing device is used to prepare and deliver water and fertilizer, enabling precise fertilization in the rice planting area and improving resource utilization efficiency; an automatic feeder is installed next to the aquaculture water body to automatically deliver feed according to the fish's growth stage and feeding strategy, reducing human intervention; soil and water quality testing equipment is deployed inside the farming unit to collect core parameters such as soil nutrients, water pH, and dissolved oxygen in real time, providing data support for control decisions. This layout, through standardized spatial design and facility configuration, realizes the engineering and replicability of the rice-fish co-culture model. Simultaneously, the supporting monitoring and control facilities provide the hardware foundation for the implementation of the intelligent system, serving as a spatial carrier integrating ecological farming and intelligent management. Figure 3 is a block diagram of the system hardware structure.

[0038] After completing the above hardware deployment and power-on debugging, the system enters an automated and intelligent closed-loop management process. As shown in Figure 8, a rice-fish integrated farming method based on multi-source data fusion and intelligent collaborative decision-making includes the following steps: S1: Synchronous acquisition of multi-source heterogeneous data. Multi-source heterogeneous data on water quality, soil, meteorology, and biological vision are synchronously acquired through a multi-source sensing network. The multi-source heterogeneous data is uploaded to the edge control unit and then synchronized to the cloud decision-making platform. For example, the system can automatically acquire data from each sensing unit according to a preset cycle. The water quality sensor reports dissolved oxygen concentration and water temperature every 5 minutes; the soil sensor reports nitrogen content and water content every 30 minutes; the meteorological station reports real-time temperature, wind speed, and air pressure change trends every 10 minutes; the video stream is continuously processed by a cloud image analysis model to output a quantitative index of fish school activity intensity.

[0039] S2: Multi-source data fusion and digital twin construction cloud decision-making platform performs preprocessing operations such as noise filtering, missing value imputation, spatiotemporal stamp alignment, and protocol unification on the received multi-source heterogeneous raw data. Then, it extracts key features reflecting the system state, generates a spatiotemporally consistent system panoramic feature dataset, and constructs a dynamic digital twin of the rice-fish co-culture system as the current state. For example, noise filtering can employ the Kalman filter algorithm; missing value imputation can use multiple imputation methods, performing multiple iterations through a regression model; and spatiotemporal alignment can use Kriging interpolation. Key features extracted can include soil nitrogen content, dissolved oxygen concentration in water, rice canopy NDVI index, and fish activity intensity index, etc.

[0040] S3: The knowledge base-driven dynamic supply and demand diagnosis cloud decision-making platform calls upon the built-in rice full growth period knowledge base and fish growth-environment response model library, combined with the real-time system status represented by the dynamic digital twin, to update the dynamic target relationship matrix and quantify the mutual influence between multiple management objectives.

[0041] Specifically, the cloud-based decision-making platform diagnoses the current state based on its built-in knowledge base of the entire rice growth cycle and a database of fish growth-environment response models. Based on the diagnostic results (e.g., identifying the rice as being in its peak tillering stage, the fish as being approximately 150 grams in size, and the dissolved oxygen level as low), the model first runs a dynamic target relation matrix update algorithm. This algorithm quantifies multiple management objectives of the system according to preset rules and the current scenario.

[0042] The dynamic target relationship matrix can be represented as: ,in, Indicates at time Management Objectives Management objectives The impact coefficient. The management objectives may include high rice yield, fish health, resource conservation, and ecological balance. For example, at this stage, the coefficient of "fish activity" on "rice health" can be considered. Set it to +0.3 (representing the gain from weed control), and set the coefficient of "oxygenation" on "fish health". Set it to +0.8.

[0043] S4: Co-operational Effect Modeling and Collaborative Strategy Planning. Based on the updated dynamic target relationship matrix and real-time deviation (i.e., the gap between the current system state and the ideal state expected by each management objective), a co-operational effect function incorporating the synergistic effect between objectives is constructed. Through a sequential decision engine with state transitions, the impact of feasible control action combinations on the system state within a future preset decision time domain is simulated. The long-term expected co-operational effect value of each action sequence is calculated, the optimal action sequence is selected, and a collaborative control strategy package is generated. Specifically, after obtaining the real-time deviation value, the weight values ​​corresponding to each objective are adjusted using this deviation value; the larger the deviation, the higher the corresponding weight value needs to be. Furthermore, the impact of different management actions on each objective can be reassessed based on this deviation—whether it is positive synergy or negative conflict—thereby updating the relationship matrix. The system control strategy package precisely specifies when, with what parameters, and what operations each execution device (i.e., the actuator) performs, and also includes time- and temperature-controlled instructions to guide fish migration between ditches and paddy fields.

[0044] The shared utility function is: ,in, Let X represent the total number of management objectives, and let X represent different control strategies. Indicates the first time when selecting control strategy X. The single-objective utility of a management objective Indicates the first The weight of each management objective, Represents the synergy coefficient. The function is used to measure the joint utility between pairwise objectives. Among them, The determination method is as follows: First, the state evolution of the system after implementing strategy X is simulated using the built-in rice growth model, fish growth model and environmental dynamics model; then, key indicators are extracted from the simulation results according to the evaluation criteria of each objective (such as rice yield, fish weight gain, resource consumption, etc.); finally, these indicators are mapped to standardized utility values ​​through a normalization function.

[0045] For details on how to implement this step, please refer to Figure 9 and its related description.

[0046] S5: The strategy package, distributed via a cloud-based decision-making platform, is sent to the edge control unit. The edge control unit's built-in programmable logic controller (PLC) parses the instructions and drives the intelligent actuators to perform operations such as irrigation, feeding, aeration, temperature control, and fish spatial management. For the control of continuous quantities such as pump speed and valve opening, a classic proportional-integral-derivative (PI-DE) control algorithm is used, forming a closed loop with sensor feedback to achieve precise positioning and adjustment.

[0047] Figure 10 is a schematic flowchart of intelligent water level control. In this invention, the water level can be automatically adjusted, oxygenation triggered, or warnings issued based on the water level requirements of rice at different growth stages and water quality (pH, dissolved oxygen) thresholds, ensuring that the farming environment is always in a state of optimal rice-fish farming. As shown in Figure 10, after the system starts, it first performs a sensor status self-check. If the sensor is abnormal, an error code is returned and the process terminates to ensure the reliability of data acquisition. If the sensor is normal, parameters such as water level, conductivity, pH, dissolved oxygen, turbidity, and water temperature are collected, and the data validity is verified. If the data is abnormal, an error code is returned and the process terminates. When the sensor status and the collected data are normal, the system enters automatic mode, which automatically determines the key growth stages of rice (e.g., before transplanting, after transplanting, tillering stage, late tillering stage, and booting stage) using the method in this invention. For each key growth stage, a corresponding water level threshold is preset (e.g., wv1 to wv10 in Figure 10), and differentiated water level control logic is executed.

[0048] When the water level is detected to be higher than the upper limit threshold for each stage, the inlet valve is closed and the outlet valve is opened to lower the water level; when the water level is detected to be lower than the lower limit threshold for each stage, the inlet valve is opened and the outlet valve is closed to raise the water level. This ensures the precise water level requirements of rice at each growth stage, while also taking into account the stability of the fish's living environment.

[0049] Based on water level control, the system monitors the pH and dissolved oxygen concentration of the water in real time. When the pH is higher or lower than the preset threshold (PH_M), an alkalinity or acidity alert is triggered. Then, when the dissolved oxygen concentration in the water is detected to be lower than the preset threshold (O2_M), the aeration equipment is automatically activated to increase oxygen levels. If the dissolved oxygen concentration in the water is not lower than the preset threshold (O2_M), all measured parameters are displayed and stored.

[0050] In certain special scenarios (e.g., equipment debugging, emergency intervention, and refined manual management), operators can directly intervene in water level and water quality parameters to ensure the stability of the rice-fish farming environment. After system startup, a sensor self-check is first performed. If a sensor malfunctions, an error code is returned and the process terminates, ensuring the reliability of data acquisition. If the sensor is normal, parameters such as water level, conductivity, pH, dissolved oxygen, turbidity, and water temperature are collected, and data validity is verified. If the data is abnormal, an error code is also returned. Only when both the sensor and data are normal does the system enter manual control mode. Users can select "Open Inlet Valve," "Close Inlet Valve," or proceed directly to the next step to achieve precise control of the water entering the aquaculture water body. Next, users can select "Open Outlet Valve," "Close Outlet Valve," or proceed directly to the next step to achieve bidirectional water level adjustment in conjunction with the inlet valve. Finally, users can select "Add Oxygen" or return directly to handle emergencies of insufficient dissolved oxygen in the water.

[0051] The aforementioned manual intervention method, as a supplement to the automatic control process, provides an emergency intervention channel to prevent environmental loss of control when the automated system fails. Moreover, it allows operators to make precise interventions based on their on-site experience, which can be adapted to complex and ever-changing planting and breeding scenarios, forming a dual guarantee mechanism of "automation as the main method and manual intervention as a supplement."

[0052] S6: Safety Closed-Loop Verification. During strategy execution, if high-risk agricultural operations such as fertilization or pesticide application are detected, a verification sub-process is automatically triggered. Upon initiation, the relevant sluice gates are forcibly locked, entering isolation mode (also known as the "safety isolation process"). Key harmful indicators in the field water are monitored frequently via a sensing network, and trend analysis is performed using a fish toxicology safety threshold database. Isolation is lifted after verifying environmental safety; otherwise, isolation is maintained and an alarm is triggered. The system calls a harmful substance monitoring model (usually a first-order degradation kinetic model) to perform high-frequency monitoring and trend prediction of ammonia nitrogen, nitrite, and other concentrations in the field water. The cloud model compares real-time data with thresholds in the fish toxicology safety threshold database and runs a safety decision logic function: the function returns "true" only if the monitored concentration remains below the safety threshold and is predicted to remain safe for a period of time. This judgment process combines statistical process control and time series prediction algorithms. Only when the safety conditions are met will the system generate a "safety recovery" command to lift the isolation. Otherwise, the system will maintain the isolation state and trigger an alarm, forming an unbreakable proactive safety closed loop.

[0053] Figure 11 is a flowchart of the water quality management process. As shown in Figure 11, after the system starts, it first performs a sensor status self-check. If the sensor is abnormal, it directly returns an error code and terminates the process to ensure the reliability of data acquisition. If the sensor is normal, it continues to collect parameters such as water level, conductivity, pH, dissolved oxygen, turbidity, and water temperature, and performs data validity verification. If the data is abnormal, it also returns an error code and terminates the process. When the sensor status and collected data are normal, the system enters automatic mode, receives decision parameters from the cloud, such as DO_target (target dissolved oxygen value), pH_min (minimum pH value), pH_max (maximum pH value), and Temp_target (target temperature value). It then performs multi-parameter collaborative diagnosis (determining the difference between the currently collected parameters and the target parameters) to comprehensively score the water quality health. If the calculated water quality health Q is not less than the preset threshold (i.e., the water quality meets the requirements), the current operating state is maintained. Conversely, if the calculated water quality health Q is less than the preset threshold (i.e., the water quality does not meet the requirements), a comprehensive control strategy is triggered, generating a collaborative execution command that drives the actuators (e.g., aerators, heaters, circulating pumps). If high-risk agricultural operations such as fertilization or pesticide application are identified during the execution of these strategies (e.g., aeration, heating), a safety verification process is initiated. The security verification process determines whether the corresponding operation is safe. If it is not safe, an alarm is triggered and manual intervention is required. If it is safe, the execution log and water quality changes are recorded, and the cloud-based digital twin of water quality is updated, awaiting the next monitoring cycle.

[0054] This law explicitly treats agricultural operations as a "system event" requiring special management and designs an automatically triggered, data-driven safety isolation and verification protocol for them. By monitoring specific harmful substances in real time and dynamically comparing them with biosafety thresholds, the system can add an intelligent software safety lock on top of hardware isolation. Only after rigorous model verification will the isolation be lifted and system linkage restored. This fundamentally eliminates the risk of fish poisoning caused by human error or delayed environmental degradation, achieving a leap from "fault protection" to "risk immunity."

[0055] S7: The model evolutionary learning collects a complete data loop of "real-time system state - collaborative control strategy - new state after execution - actual production utility," which can be stored as a quadruple in the cloud experience pool. The system periodically uses this data to incrementally train and optimize the parameters of the rice-fish multi-source data fusion growth decision model on the cloud decision platform through deep reinforcement learning algorithms, thereby improving its long-term utility prediction capability. This enables the model to continuously and autonomously evolve in specific production environments, making its decisions increasingly accurate and efficient.

[0056] Figure 9 is a schematic flowchart of collaborative strategy planning. As shown in Figure 9, collaborative strategy planning may include the following steps: S41: Obtain the current state of the paddy field ecosystem. That is, a dynamic digital twin, which represents the current state. Input a sequential decision engine with state transition functionality to formalize the future rice-fish farming management problem into a Markov decision process with a finite time domain; S42: Define the core elements of the Markov decision process, including the preset decision time domain, state space, and action space. and state transition function, wherein the decision time domain is set to 24 hours, and the action space The state transition function, defined by the system's built-in system dynamic model, describes the execution actions and includes all adjustable variables and their value ranges, such as irrigation amount, feeding amount, aeration intensity, and gate opening. After state from Transfer to The probability and outcome; S43: in the current state As the root node of the strategy tree, from the action space Extract all possible actions in the current state, and use each possible action as a branch edge of the root node. Simulate the next state after each action is executed through the state transition function. S44: Split the decision time domain according to the preset decision step size, recursively expand the policy tree, and for nodes that have not been expanded in the policy tree, extract the possible actions in the corresponding state and simulate the state transition to generate child nodes in turn, until the expansion depth of all nodes covers the entire decision time domain, forming a policy tree containing all possible action sequences. Each path from the root node to the leaf node corresponds to a set of continuous control action sequences in the decision time domain; S45: Use a simulation-based search algorithm to traverse and evaluate the paths in the policy tree through four iterations: selection, expansion, simulation, and backtracking, and calculate the long-term expected synergistic effect of each path. ,in, , This represents the total number of decision steps broken down within the decision-making time domain. Indicates the discount factor. Indicates the first The prediction co-effect in the step state is achieved by Calculated; S46: Selected The path with the highest value is the optimal coordinated control action sequence within the decision-making time domain; S47: The optimal coordinated control action sequence is packaged into a coordinated control strategy package through a strategy encapsulation algorithm. The coordinated control strategy package specifies the timing operation parameters of each intelligent actuator. The operation parameters include: execution time, control intensity, and runtime. The coordinated control strategy package also includes timing and temperature control instructions to guide fish to migrate directionally between aquaculture ditches and paddy fields, ensuring that the actions of each intelligent actuator are coordinated without conflict and meet the needs of rice-fish symbiosis.

[0057] Furthermore, this system supports modular expansion, allowing for easy integration with new sensing devices such as drones and sonar monitoring, or data exchange with third-party farm management information systems. The system possesses a robust self-diagnosis and fault-tolerance mechanism. When abnormal sensor data, communication interruptions, or actuator malfunctions are detected, it can automatically trigger local alarms, activate backup strategies, and report fault information via remote communication. It also supports remote OTA upgrades to fix software vulnerabilities or update decision model parameters, ensuring long-term stable operation of the system.

[0058] In summary, the specific implementation of this invention, through standardized hardware deployment and programmed intelligent model operation, realizes a complete closed loop from physical perception, state recognition, risk prediction, collaborative regulation to self-optimization, providing a systematic solution for the precise and intelligent management of rice-fish integrated farming.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A rice-fish integrated farming system based on multi-source data fusion and intelligent collaborative decision-making, characterized in that, include: The system comprises engineered paddy field facilities, a multi-source sensing network, intelligent actuators, an edge control unit, and a cloud-based decision-making platform. The engineered paddy field facilities include aquaculture ditches, controllable inlets and outlets, paddy field planting areas, and controllable interconnecting inlets between the aquaculture ditches and the paddy field planting areas. The multi-source sensing network is deployed in and around the paddy field to simultaneously collect heterogeneous data on water quality, soil, meteorology, and biological vision. The intelligent actuators include oxygenation, heating, feeding, and water circulation control equipment, which execute rice-fish farming production actions according to control commands issued by the edge control unit. The edge control unit has a built-in programmable logic controller (PLC) that connects to the multi-source sensing network and intelligent actuators via a local communication network to achieve data aggregation, command parsing, and real-time control, and interacts with the cloud-based decision-making platform via a remote communication network. A cloud-based decision-making platform deploys a rice-fish multi-source data fusion decision-making model. This model incorporates a rice full-growth-cycle knowledge base and a fish growth-environment response model library. Utilizing the rice-fish multi-source data fusion decision-making model and sensory data collected by a multi-dimensional sensing network, it generates a collaborative regulation strategy that balances the synergistic effect of rice-fish symbiosis with long-term production benefits. The collaborative regulation strategy is then distributed to the edge control unit, enabling the edge control unit to drive the intelligent actuator to execute the collaborative regulation strategy and receive feedback data after strategy execution to complete the model's autonomous optimization.

2. The rice-fish integrated farming system based on multi-source data fusion and intelligent collaborative decision-making according to claim 1, characterized in that, The aquaculture ditch surrounds the paddy field planting area and has multiple controllable interconnecting water inlets, which are spaced apart along the side wall of the aquaculture ditch. Each interconnecting water inlet is equipped with a flat gate that can be raised and lowered by a motor, and the opening degree and opening time of the flat gate are controllable.

3. The rice-fish integrated farming system based on multi-source data fusion and intelligent collaborative decision-making according to claim 1, characterized in that, The multi-source sensing network includes a water quality monitoring module, a soil monitoring module, a meteorological monitoring module, and a visual monitoring module. The water quality monitoring module is deployed at points with stable water flow in the aquaculture ditches to monitor dissolved oxygen concentration, pH value, conductivity, water temperature, and water level changes in the ditches. The soil monitoring module is deployed in the paddy field planting area to monitor soil volumetric water content, temperature, pH value, conductivity, and nutrient content. The meteorological monitoring module monitors air temperature, relative humidity, wind speed, wind direction, light intensity, rainfall, and atmospheric pressure in the fields. The visual monitoring module acquires visual data from the aquaculture ditches and paddy fields.

4. The rice-fish integrated farming system based on multi-source data fusion and intelligent collaborative decision-making according to claim 1, characterized in that, The intelligent actuator includes an ultrafine bubble aeration pipe network and a heating pipe network laid at the bottom of the aquaculture ditch, an automatic feeder installed at the edge of the aquaculture ditch, and a water circulation control system consisting of an irrigation pump, a circulation pump, and an electric inlet and outlet valve.

5. A method for integrated rice-fish farming based on multi-source data fusion and intelligent collaborative decision-making, employing the system described in any one of claims 1-4, characterized in that, Includes the following steps: S1: Synchronous Acquisition of Multi-Source Heterogeneous Data: Multi-source heterogeneous data on water quality, soil, meteorology, and biological vision are synchronously acquired through a multi-source sensing network. This multi-source heterogeneous data is uploaded to the edge control unit and then synchronized to the cloud decision-making platform. S2: Multi-Source Data Fusion and Digital Twin Construction: The cloud decision-making platform preprocesses the received multi-source heterogeneous raw data and extracts key features reflecting the system state, generating a spatiotemporally consistent system panoramic feature dataset to construct a dynamic digital twin of the rice-fish co-culture system. S3: Knowledge Base-Driven Dynamic Supply and Demand Diagnosis: The cloud decision-making platform calls upon the built-in rice full-growth-cycle knowledge base and fish growth-environment response model library, combining the real-time system state represented by the dynamic digital twin to update the dynamic target relationship matrix and quantify the mutual influence between multiple management objectives. S4: Co-operation modeling and collaborative strategy planning. Based on the updated dynamic target relationship matrix, a co-operation function incorporating the synergistic effect between targets is constructed. Through a sequential decision engine with state transition, the impact of feasible control action combinations on the system state within the future preset decision time domain is simulated. The long-term expected co-operation value of each action sequence is calculated, the optimal action sequence is selected, and a collaborative control strategy package is generated. S5: The strategy package for coordinated control of strategy distribution and precise execution is distributed from the cloud decision-making platform to the edge control unit. After the programmable logic controller built into the edge control unit parses the instructions, it drives the intelligent actuator to complete operations such as irrigation, feeding, oxygenation, temperature regulation and fish space scheduling. S6: Safety closed-loop verification During the strategy execution process, if high-risk agricultural operations such as fertilization or pesticide application are identified, the safety isolation process is automatically triggered. Key harmful indicators in the field water are monitored at high frequency through the sensing network, and trend analysis is performed in combination with the fish toxicology safety threshold database. After verifying that the environment is safe, the isolation is lifted; otherwise, the isolation is maintained and an alarm is triggered. S7: Model evolutionary learning collects a complete data loop of "real-time system state - collaborative control strategy - new state after execution - actual production utility". Based on this data loop, incremental training and parameter optimization are performed on the rice-fish multi-source data fusion growth decision model of the cloud decision platform to achieve autonomous model evolution.

6. The rice-fish integrated farming method based on multi-source data fusion and intelligent collaborative decision-making according to claim 5, characterized in that, The preprocessing includes noise filtering, missing value interpolation, timestamp alignment, and protocol unification.

7. The rice-fish integrated farming method based on multi-source data fusion and intelligent collaborative decision-making according to claim 5, characterized in that, Key features extracted included soil nitrogen content, dissolved oxygen concentration in water, NDVI index of rice canopy, and fish activity intensity index.

8. The rice-fish integrated farming method based on multi-source data fusion and intelligent collaborative decision-making according to claim 5, characterized in that, The dynamic target relationship matrix is: ,in, Indicates at time Management Objectives Management objectives The impact coefficient is determined by multiple management objectives, including high rice yield, fish health, resource conservation, and ecological balance.

9. The rice-fish integrated farming method based on multi-source data fusion and intelligent collaborative decision-making according to claim 8, characterized in that, The shared utility function is: ,in, Let X represent the total number of management objectives, and let X represent different control strategies. Indicates the first time when selecting control strategy X. The single-objective utility of a management objective Indicates the first The weight of each management objective, Represents the synergy coefficient. The function is used to measure the joint utility between two pairs of objectives.

10. The rice-fish integrated farming method based on multi-source data fusion and intelligent collaborative decision-making according to claim 9, characterized in that, S4 specifically includes: S41: Obtaining the current state of the paddy field ecosystem. That is, a dynamic digital twin, which represents the current state. Input a sequential decision engine with state transition functionality to formalize the future rice-fish farming management problem into a Markov decision process with a finite time domain; S42: Define the core elements of the Markov decision process, including the preset decision time domain, state space, and action space. and state transition function, wherein the decision time domain is set to 24 hours, and the action space The state transition function, defined by the system's built-in system dynamic model, describes the execution actions and includes all adjustable variables and their value ranges, such as irrigation amount, feeding amount, aeration intensity, and gate opening. After state from Transfer to The probability and outcome; S43: in the current state As the root node of the strategy tree, from the action space Extract all possible actions in the current state, and use each possible action as a branch edge of the root node. Simulate the next state after each action is executed through the state transition function. S44: Split the decision time domain according to the preset decision step size, recursively expand the policy tree, and for nodes that have not been expanded in the policy tree, extract the possible actions in the corresponding state and simulate the state transition to generate child nodes in turn, until the expansion depth of all nodes covers the entire decision time domain, forming a policy tree containing all possible action sequences. Each path from the root node to the leaf node corresponds to a set of continuous control action sequences in the decision time domain; S45: Use a simulation-based search algorithm to traverse and evaluate the paths in the policy tree through four iterations: selection, expansion, simulation, and backtracking, and calculate the long-term expected synergistic effect of each path. ,in, ， This represents the total number of decision steps broken down within the decision-making time domain. Indicates the discount factor. Indicates the first The prediction co-effect in the step state is achieved by Calculated; S46: Selected The path with the highest value is the optimal coordinated control action sequence within the decision-making time domain; S47: The optimal coordinated control action sequence is packaged into a coordinated control strategy package through a strategy encapsulation algorithm. The coordinated control strategy package specifies the timing operation parameters of each intelligent actuator. The operation parameters include: execution time, control intensity, and runtime. The coordinated control strategy package also includes timing and temperature control instructions to guide fish to migrate directionally between aquaculture ditches and paddy fields, ensuring that the actions of each intelligent actuator are coordinated without conflict and meet the needs of rice-fish symbiosis.

Images