Cloud edge end collaborative fruit and vegetable water and fertilizer intelligent management and control method and system

Abstract

The application provides a fruit and vegetable water and fertilizer intelligent management and control method and system based on cloud edge end cooperation, the method comprises the following steps: an edge device determines a planting site state vector according to multi-source detection data of a fruit and vegetable planting site; based on the planting site state vector, a basic water and fertilizer strategy is optimized in real time, and a water and fertilizer control instruction is sent to an end-side device; the basic water and fertilizer strategy is generated based on a water and fertilizer control model trained by a cloud end; execution state feedback data sent by the end-side device is received; in the case that a network is disconnected between the edge device and the cloud end, based on the state feedback data and the cached water and fertilizer control model, network disconnection autonomy of fruit and vegetable water and fertilizer management and control is carried out. Through cloud edge end cooperation, the application effectively reduces the artificial experience threshold while ensuring the accuracy of water and fertilizer regulation, and provides a smart water and fertilizer management and control scheme for large-scale facility planting of multiple sheds, multiple crops and multiple devices.

Description

Technical Field

[0001] This invention relates to the fields of smart agriculture, agricultural Internet of Things, edge computing, cloud platform operation and maintenance, and integrated water and fertilizer control technology, and in particular to a method and system for intelligent management and control of water and fertilizer for fruits and vegetables based on cloud-edge-device collaboration. Background Technology

[0002] In recent years, people have paid increasing attention to food safety. As an important part of daily diet, the water and fertilizer management of fruits and vegetables directly affects their healthy growth and quality. With the continuous development of large-scale facility cultivation of specialty fruits and vegetables, planting areas typically encompass multiple greenhouses, various crops, different types of water and fertilizer equipment, and sensors from different manufacturers. While traditional water and fertilizer management solutions can reduce the labor intensity of irrigation and fertilization to some extent, the high operational threshold of the equipment requires not only experienced professionals but also intelligent management methods. Given the critical impact of water and fertilizer management on the quality and safety of fruits and vegetables, establishing a comprehensive water and fertilizer management system for fruits and vegetables is particularly urgent. Summary of the Invention

[0003] This invention provides a method and system for intelligent water and fertilizer management of fruits and vegetables based on cloud-edge-device collaboration. Through cloud-edge-device collaboration, while ensuring the accuracy of water and fertilizer regulation, it effectively reduces the threshold of human experience and provides an intelligent water and fertilizer management solution for large-scale facility planting with multiple greenhouses, multiple crops, and multiple equipment.

[0004] This invention provides a cloud-edge-device collaborative intelligent water and fertilizer management method for fruits and vegetables, applied to edge devices. The method includes: determining a planting site state vector based on multi-source detection data from the fruit and vegetable planting site; performing real-time rolling optimization of a basic water and fertilizer strategy based on the planting site state vector, and sending water and fertilizer control commands to the edge device; the basic water and fertilizer strategy is generated based on a water and fertilizer control model trained in the cloud; receiving execution status feedback data sent by the edge device; and performing autonomous water and fertilizer management of fruits and vegetables in the event of a network outage between the edge device and the cloud, based on the status feedback data and the cached water and fertilizer control model.

[0005] According to the present invention, a cloud-edge-device collaborative intelligent management and control method for water and fertilizer in fruits and vegetables is provided. The step of determining the state vector of the planting site based on multi-source detection data of the fruit and vegetable planting site includes: performing protocol adaptation and data cleaning and fusion on the multi-source detection data of the fruit and vegetable planting site to obtain the state vector of the planting site; the fields of the state vector of the planting site include one or more combinations of timestamp, greenhouse number, soil moisture content, soil conductivity, soil temperature, air temperature, relative humidity, light intensity, fertilizer solution conductivity, fertilizer solution pH, pipeline flow rate, pipeline pressure and fertilizer solution level.

[0006] According to the present invention, a cloud-edge-device collaborative intelligent water and fertilizer management method for fruits and vegetables is provided. The method involves real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector, and sending water and fertilizer control commands to the end-side device. This includes: inputting the planting site state vector and the target irrigation and fertilizer injection amounts from the basic water and fertilizer strategy into a water and fertilizer execution process prediction model to obtain the pump and valve control quantities output by the prediction model; and sending the water and fertilizer control commands to the end-side device based on the pump and valve control quantities. The water and fertilizer execution process prediction model aims to minimize the deviation between the predicted planting state and the target planting state, and updates the effective accumulated temperature during the water and fertilizer execution process. Constraints include root zone salt accumulation constraints.

[0007] According to the present invention, a cloud-edge-device collaborative intelligent management method for water and fertilizer in fruits and vegetables is provided. The method, based on the state feedback data and a cached water and fertilizer control model, performs autonomous water and fertilizer management of fruits and vegetables offline. This includes: when the cached water and fertilizer control model and the corresponding water and fertilizer strategy are within their validity period, using the cached water and fertilizer control model and the state feedback data to perform rolling optimization and water and fertilizer control of the water and fertilizer strategy; or, when the cached water and fertilizer control model is valid and the validity period of the water and fertilizer strategy has expired, based on the state feedback data and the cached water and fertilizer control model, reverting the parameters of the water and fertilizer strategy to a preset target water and fertilizer regulation value; or, when the cached water and fertilizer control model has exceeded the validity period of the water and fertilizer strategy, controlling the water and fertilizer strategy to enter a nutrient stress buffer control mode; the nutrient stress buffer control mode is used to regulate the water and fertilizer strategy according to a preset load reduction rate.

[0008] The intelligent management and control method for water and fertilizer in fruits and vegetables based on cloud-edge-device collaboration provided by the present invention further includes: determining the status of the fruit and vegetable planting site according to the execution status feedback data; executing safety interlock control when the status of the fruit and vegetable planting site is abnormal; the abnormal status of the fruit and vegetable planting site includes one or more combinations of pipeline pressure exceeding the limit, fertilizer liquid level below the lower limit, sensor offline, actuator feedback abnormality, and communication link abnormality; the safety interlock control includes one or more combinations of load reduction operation, shutting down the fertilizer pump, shutting down the abnormal branch solenoid valve, and stopping the water supply pump.

[0009] The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration provided by the present invention further includes: generating the water and fertilizer control instructions based on the rolling optimized water and fertilizer strategy; wherein the rolling optimized water and fertilizer strategy is a strategy verified by digital twin simulation.

[0010] This invention also provides a cloud-edge-device collaborative intelligent management method for water and fertilizer in fruits and vegetables, applied in the cloud. The method includes: aggregating local water and fertilizer models sent by edge devices in each greenhouse area through federated learning to generate a global model; training the model based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; generating a basic water and fertilizer strategy based on the trained water and fertilizer control model, and sending the basic water and fertilizer strategy to the edge devices.

[0011] The present invention provides a cloud-edge-device collaborative intelligent management method for water and fertilizer in fruits and vegetables, which further includes: performing digital twin simulation verification on the rolling optimized water and fertilizer strategy sent by the edge device in a digital twin fruit and vegetable planting environment; the rolling optimized water and fertilizer strategy is a strategy after real-time rolling optimization of the basic water and fertilizer strategy; and sending the simulated and verified water and fertilizer strategy to the edge device.

[0012] This invention also provides a cloud-edge-device collaborative intelligent water and fertilizer management system for fruits and vegetables, applied to edge devices. The system includes: a first module for determining a planting site state vector based on multi-source detection data from the fruit and vegetable planting site; a second module for performing real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector and sending water and fertilizer control commands to the edge device; the basic water and fertilizer strategy is generated based on a water and fertilizer control model trained in the cloud; a third module for receiving execution status feedback data sent by the edge device; and a fourth module for performing autonomous water and fertilizer management of fruits and vegetables in the event of a network outage between the edge device and the cloud, based on the status feedback data and the cached water and fertilizer control model.

[0013] This invention also provides a cloud-edge-device collaborative intelligent water and fertilizer management system for fruits and vegetables, applied in the cloud. The system includes: a fifth module, used to aggregate local water and fertilizer models sent by edge devices in each greenhouse area through federated learning to generate a global model; a sixth module, used to train the model based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; and a seventh module, used to generate a basic water and fertilizer strategy based on the trained water and fertilizer control model and send the basic water and fertilizer strategy to the edge devices.

[0014] This invention provides a cloud-edge-device collaborative intelligent water and fertilizer management method and system for fruits and vegetables. Through cloud-edge-device collaboration, while ensuring the accuracy of water and fertilizer regulation, it effectively reduces the threshold of human experience and provides an intelligent water and fertilizer management solution for large-scale facility planting with multiple greenhouses, multiple crops, and multiple equipment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is one of the flowcharts of an intelligent management and control method for water and fertilizer of fruits and vegetables based on cloud-edge-device collaboration provided by the present invention.

[0017] Figure 2 This is the second flowchart of a cloud-edge-device collaborative intelligent management and control method for water and fertilizer in fruits and vegetables provided by the present invention.

[0018] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] Please refer to Figure 1 , Figure 1 This is one of the flowcharts illustrating an intelligent management and control method for water and fertilizer in fruits and vegetables based on cloud-edge-device collaboration provided by the present invention.

[0021] This invention provides a cloud-edge-device collaborative intelligent management method for water and fertilizer control of fruits and vegetables, applied to edge devices. The method includes: 101: Determine the state vector of the planting site based on multi-source detection data from the fruit and vegetable planting site; 102: Based on the state vector of the planting site, the basic water and fertilizer strategy is optimized in real time and water and fertilizer control commands are sent to the end device; the basic water and fertilizer strategy is generated based on the water and fertilizer control model trained in the cloud. 103: Execution status feedback data sent by the receiving end-side device; 104: In the event of a network outage between edge devices and the cloud, autonomous management of water and fertilizer for fruits and vegetables is achieved based on status feedback data and a cached water and fertilizer control model.

[0022] The end-side equipment in this invention is deployed at the planting site and includes a soil sensor group (soil moisture sensor, soil electrical conductivity (EC) sensor, soil temperature sensor), a meteorological sensor group (air temperature and humidity sensor, light intensity sensor), a water and fertilizer pipeline sensor group (fertilizer solution EC sensor, fertilizer solution pH sensor, pipeline flow sensor, pipeline pressure sensor, fertilizer solution level sensor), a water supply pump, a fertilizer pump, a solenoid valve, and an alarm actuator. The end-side equipment is used to collect multi-source detection data from the fruit and vegetable planting site and execute water and fertilizer control commands.

[0023] Edge devices are deployed on local servers or edge gateways within the plantation, with the edge gateways communicating with both the edge devices and the cloud. The edge devices clean, align, and fuse the raw data uploaded by the edge devices to form a standardized plantation site status vector.

[0024] Deployed on a public or private cloud platform, the cloud trains or updates the water and fertilizer control model based on historical operation data and environmental data uploaded by each greenhouse area, generates basic water and fertilizer strategies, and distributes the model parameters, strategies, and version numbers to edge devices.

[0025] Based on the state vector of the planting site, the edge device performs real-time (e.g., minute-level or second-level) rolling optimization of the basic water and fertilizer strategy, and generates water and fertilizer control instructions (pump and valve execution instructions) to be sent to the end device.

[0026] The edge device will execute feedback and return it to the edge device in real time.

[0027] Given the numerous real-time control needs at facility planting sites (such as emergency shutdowns during sudden pressure surges), relying entirely on cloud-based decision-making would lead to untimely safety responses due to end-to-end communication delays and network unavailability, resulting in equipment damage or production accidents.

[0028] When communication between the edge device and the cloud fails (e.g., network disconnection or heartbeat timeout), the edge device continues to perform local rolling optimization based on the cached effective model and strategy, and caches running data in chronological order through an offline task queue to achieve autonomous control during network outages. After the network is restored, the edge device re-uploads the offline data to the cloud. The cloud performs consistency verification based on the model version number.

[0029] This embodiment enables the system to maintain continuous operation of water and fertilizer control even when the network is abnormal by dividing the functions of the cloud, edge, and terminal layers and decoupling the time scale, while ensuring the consistency and traceability between cloud model updates and edge execution.

[0030] As a preferred embodiment, the planting site state vector is determined based on multi-source detection data from the fruit and vegetable planting site. This includes: performing protocol adaptation and data cleaning and fusion on the multi-source detection data from the fruit and vegetable planting site to obtain the planting site state vector. The fields of the planting site state vector include one or more combinations of timestamp, greenhouse number, soil moisture content, soil conductivity, soil temperature, air temperature, relative humidity, light intensity, fertilizer solution conductivity, fertilizer solution pH, pipeline flow rate, pipeline pressure, and fertilizer solution level.

[0031] Considering that the sensors and actuators within the planting area come from different manufacturers and support different communication protocols, this embodiment supports protocol adaptation for edge devices, such as at least two of the following communication protocols or methods: Modbus Communication Protocol (Modbus), Recommended Standard 485 (RS485), Controller Area Network (CAN), Long Range Low Power Wireless Communication Technology (LoRa), Narrow Band Internet of Things (NB-IoT), 4th Generation Mobile Communication Technology (4G), 5th Generation Mobile Communication Technology (5G), Wireless Fidelity (WiFi), and Ethernet. A unified data object is generated for sensors and actuators from different manufacturers and of different models, converting the raw data reported by each device (such as different register addresses, data formats, units, and sampling frequencies) into a standardized format.

[0032] For example, in a tomato growing facility, soil moisture sensors report volumetric moisture content using RS485 / Modbus protocol, soil moisture sensors report soil humidity using LoRa wireless communication, and fertilizer solution EC sensors report conductivity using 4G communication. The edge device's protocol adaptation module loads corresponding protocol parsing plugins for these three types of devices, uniformly converting the raw data into standardized data objects, including device number, data item name (uniformly soil moisture content and fertilizer solution EC value), numerical value (uniformly converted to standard units), timestamp, and quality label.

[0033] When the park expands to new areas and adds equipment from new manufacturers, it is only necessary to load the corresponding protocol parsing plugin or configuration file into the protocol adaptation module. Once the new equipment is online, it automatically registers and begins reporting data, and the protocol adaptation module automatically completes the data format conversion. Through standardized conversion by the protocol adaptation module, plug-and-play functionality for heterogeneous devices is achieved, significantly reducing system integration costs and the workload of debugging during equipment replacement.

[0034] Edge devices include a data cleaning and fusion module, which is used to preprocess multi-source sensor data reported by edge devices. The preprocessing methods include timestamp alignment, outlier removal, missing value imputation, and multi-point spatial fusion.

[0035] Timestamp alignment: Since different sensors have different sampling frequencies (e.g., soil sensors sample once every 10 seconds, and meteorological sensors sample once every 30 seconds), the data streams are interpolated and aligned to the same time reference point based on a preset time window (e.g., 1 second or 5 seconds).

[0036] Outlier removal: Set reasonable value ranges for each data item (e.g., soil moisture content 0-60%, air temperature -20-50℃, fertilizer solution EC value 0-10mS / cm). Data outside the range is judged as outliers and removed. At the same time, the sliding window method is used to identify sudden jumps (e.g., pressure jumps from 0.2MPa to 1.0MPa within 1 second) and mark them as suspicious data for subsequent processing.

[0037] Missing value imputation: For missing data caused by temporary sensor offlineness or communication packet loss, linear interpolation, historical average, or spatial interpolation of nearby sensors are used to imput the missing data. If a soil moisture sensor has missing data for no more than 3 consecutive sampling periods, linear interpolation is used to imput the missing data; if it has missing data for more than 3 periods, it is marked as continuously missing and the sensor offline alarm is triggered.

[0038] Multi-point spatial fusion: When multiple sensors of the same type are deployed in the same greenhouse area (such as four soil moisture sensors arranged in different locations), the measured values ​​of each point are spatially fused. The representative value of the greenhouse area can be obtained by using the average value, median value, or weighted average based on the historical reliability of each point.

[0039] The data cleaning and fusion module aggregates all the preprocessed data into a standardized planting site status vector, which includes fields such as timestamp, greenhouse number, soil moisture content, soil EC value, soil temperature, air temperature, relative humidity, light intensity, fertilizer solution EC value, fertilizer solution pH value, pipeline flow rate, pipeline pressure, and fertilizer solution level.

[0040] This embodiment effectively solves the problems of inconsistent quality, asynchronous time, and uneven spatial distribution of field sensor data through data cleaning and fusion.

[0041] Given that cloud-based models and edge execution are prone to version inconsistencies, if the model is updated in the cloud but not promptly synchronized to the edge devices, or if the edge devices continue to use outdated policies after network recovery, it will lead to a mismatch between control commands and the actual needs of crops on site, increasing production risks.

[0042] In this embodiment, the edge device includes a model caching module, which stores model-related information distributed from the cloud, such as model files (containing parameter matrices of control algorithms or neural network weights), model version numbers (e.g., V2.3.1), water and fertilizer strategies (including target moisture content ranges, target EC values, nitrogen, phosphorus, and potassium ratios, pulse parameters, etc. for each growth stage), strategy validity periods (e.g., from June 1, 2025 to July 15, 2025), applicable crops, and applicable growth stages. The model caching module adopts a circular buffer design with a preset maximum storage capacity of 50,000 records. When the limit is reached, expired records are deleted in ascending order.

[0043] Version number naming conventions can adopt a semantic version number format: major version number.minor version number.revision number (e.g., V2.3.1). The major version number increments when the crop model architecture changes or the core algorithm is replaced (e.g., switching from MPC to reinforcement learning); the minor version number increments when the model parameters are updated on a large scale, or when new crop varieties or growth stages are added; the revision number increments when policy parameters are fine-tuned, bugs are fixed, or thresholds are adjusted.

[0044] When a new water and fertilizer control model is distributed from the cloud, it includes a version number. The edge device receives the model packet (containing the model file, version number, and Message Digest Algorithm 5 (MD5)), calculates its local MD5 hash, compares it with the MD5 hash distributed from the cloud, and rejects the packet if they do not match, reporting the discrepancy. After successful model verification, the edge device marks the current cached model as a historical version, writes a new model record, updates its local model version, and sends a confirmation message (including the new version number and local update time) back to the cloud.

[0045] After network recovery, model version consistency verification is required. Specifically, edge devices report their current local model version to the cloud. The cloud compares this with the latest global version number. If they match, offline data is re-uploaded normally; otherwise, the cloud distributes the latest model, and the edge devices perform a forced update. When the major version number changes, a full update of the model files is forced; when the minor version number or revision number changes, only the difference parameters are updated. After the update is complete, the edge devices re-upload offline data (labeled with the old version number) in chronological order. The cloud archives the old version data separately, without mixing it into the new version training set.

[0046] As a preferred embodiment, based on the planting site state vector, the basic water and fertilizer strategy is optimized in real time and then water and fertilizer control commands are sent to the end-side equipment. This includes: inputting the planting site state vector and the target irrigation amount and target fertilizer injection amount in the basic water and fertilizer strategy into the water and fertilizer execution process prediction model, obtaining the pump valve control amount output by the water and fertilizer execution process prediction model, and sending water and fertilizer control commands to the end-side equipment based on the pump valve control amount; the water and fertilizer execution process prediction model takes the minimum deviation between the predicted planting state and the target planting state as the optimization objective, and updates the effective accumulated temperature during the water and fertilizer execution process, with constraints including root zone salt accumulation constraints.

[0047] In this embodiment, the edge device includes a real-time optimization module, which embeds a water and fertilizer execution process prediction model with root zone salt accumulation constraints. Within each control cycle, the module solves the problem of minimizing the following cost function: , in, y The output vector contains the root zone water content and the root zone EC value; y ref The dynamic reference trajectory of the cloud strategy after temperature accumulation correction; To control the increments, we include the increments in water supply pump frequency, fertilizer pump speed, and solenoid valve opening. Q and R This is the weight matrix. For relaxation factor weights, These are soft-constraint slack variables; i It is the i-th day from the date of sowing or transplanting. k Np is the sequence number of the current control cycle, Nc is the prediction time domain, and Nc is the control time domain.

[0048] Weight matrix settings (using tomatoes as an example): in, Q The high weight of medium moisture content (2.0) reflects the priority of water in ensuring fruit expansion in tomatoes; R The duty cycle has a relatively high weight (0.5) to suppress frequent operation of the solenoid valve and extend its service life.

[0049] Hard constraints include upper limits on pipeline pressure. fertilizer solution lower limit and the limit of EC change rate in the root zone ; The constraint on the EC change rate in the root zone aims to prevent root salt burn caused by excessively rapid fertilization.

[0050] When the local model expires due to policy expiration and the network is abnormal: in t exp When the strategy expires, T buffer For the buffer period (e.g., 7 days). The unloading rate coefficient (taken as 0.05 / day) is the final... It remains constant after decreasing to 70% of the initial value.

[0051] The real-time optimization module is set with a rolling optimization cycle of Ts=60, a prediction time domain of Np=10 (i.e., the next 10 minutes), and a control time domain of Nc=3 (i.e., the control quantity in the first 3 minutes participates in the optimization, and then remains constant).

[0052] Effective accumulated temperature GDD Updated daily: , in T base Where n is the baseline growth temperature for crops (e.g., 10℃ for tomatoes), and n is the total sample size. Let the maximum temperature of day i be _____. Let be the minimum temperature of day i.

[0053] Accumulated temperature correction function of reference trajectory for: , Among them, when hour, ; when hour, ; when hour, , The target moisture content in the basic strategy distributed from the cloud. The target moisture content for the next reproductive stage. This represents the initial threshold for accumulated temperature during the current reproductive stage. This represents the accumulated temperature threshold for the current reproductive stage. The accumulated temperature progress factor (0~1) represents the relative position of the current GDD within the phase switching interval.

[0054] When GDD crosses the stage switching threshold, a smooth transition factor is introduced. Avoid target jumps: , in, , This is the transition time constant (taken as 1800 seconds). To control the cycle.

[0055] At each time step, a simplified linear incremental prediction model is established as follows: Wherein, the state vector , representing root zone moisture content (%), root zone EC value (mS / cm), pipeline pressure (MPa), fertilizer solution level (m), and soil temperature (°C), respectively; control vector , respectively represent the water supply pump frequency (Hz), fertilizer pump speed (rpm), and solenoid valve opening and closing duty cycle (0~1); w and v are the process noise and measurement noise, respectively.

[0056] Typical parameter matrices obtained through system identification (taking greenhouse tomatoes as an example): The rolling optimization cycle of edge devices is shorter than the model update cycle of the cloud. Edge devices correct pump and valve parameters on a minute-level (e.g., 1 minute, 5 minutes) or second-level (e.g., 30 seconds) cycle to respond to changes in field conditions (e.g., increased evaporation due to a sudden increase in sunlight, pressure fluctuations in a certain branch, etc.).

[0057] The cloud updates model parameters and basic strategies on a daily (e.g., updating the strategy for the next day at midnight every day), weekly (e.g., updating the strategy for the week every Monday), or crop growth season (e.g., updating strategy parameters at the beginning of each growth stage) basis.

[0058] This embodiment ensures both the optimization and iteration of long-term strategies and a rapid response to short-term disturbances.

[0059] As a preferred embodiment, offline autonomous management of water and fertilizer control for fruits and vegetables is performed based on state feedback data and a cached water and fertilizer control model. This includes: when the cached water and fertilizer control model and the corresponding water and fertilizer strategy are within their validity period, using the cached water and fertilizer control model and state feedback data to perform rolling optimization and water and fertilizer control of the water and fertilizer strategy; or, when the cached water and fertilizer control model is valid and the validity period of the water and fertilizer strategy has expired, based on the state feedback data and the cached water and fertilizer control model, reverting the parameters of the water and fertilizer strategy to a preset target water and fertilizer regulation value; or, when the cached water and fertilizer control model has exceeded the validity period of the water and fertilizer strategy, controlling the water and fertilizer strategy to enter a nutrient stress buffer control mode; the nutrient stress buffer control mode is used to regulate the water and fertilizer strategy according to a preset load reduction rate.

[0060] In this embodiment, the edge device maintains periodic heartbeat detection with the cloud (e.g., once every 30 seconds). When the edge device fails to receive a heartbeat response from the cloud multiple times consecutively (e.g., 3 times), it determines that the communication link is abnormal and automatically enters the network disconnection autonomous mode.

[0061] In the offline autonomous mode, the edge device checks whether the model and strategy corresponding to the current crop and growth stage in the model cache module are still valid. If they are valid, the model continues to be used for local rolling optimization, and water and fertilizer control is unaffected.

[0062] If the local model is still valid but the policy is about to expire (e.g., less than 24 hours remaining), the edge device will automatically slightly reduce the parameters within the policy's validity period to a conservative value (e.g., reduce the target EC value by 10%) to address the risk that the network will not recover after the policy expires.

[0063] If the local model has exceeded the policy validity period and the network has not yet recovered, the edge device enters the nutrient stress buffer control mode: the target fertilizer solution EC value is gradually reduced to 70% of the initial value at the preset load reduction rate and then kept constant, the number of fertilization pulses is reduced or the fertilization interval is extended, and crop safety is prioritized over optimal growth efficiency until the network is restored or manual intervention is required.

[0064] For example, in a melon cultivation facility, a model version V3.1 for the fruit expansion stage was distributed from the cloud, with a policy validity period from June 1, 2025 to June 30, 2025. On June 15, a network outage occurred, but the policy remained valid in the model cache module, and edge devices continued to perform local optimization according to V3.1. On June 28, with the network still not restored and the policy about to expire, the edge devices automatically and gradually reduced the target EC value from 2.2 mS / cm to 1.8 mS / cm. On July 1, the network was restored, and the edge devices retransmitted all data from the outage period and requested the latest model.

[0065] This embodiment ensures that the system has a basis for action and a strategy to follow during network outages through model caching, version number management, and policy validity mechanisms. It avoids system shutdown or blind decision-making when the model is missing, while ensuring version consistency between cloud model updates and edge execution.

[0066] As a preferred embodiment, the method further includes: determining the on-site status of fruit and vegetable planting based on execution status feedback data; executing safety interlock control in the event of an abnormal on-site status of fruit and vegetable planting; the abnormal on-site status of fruit and vegetable planting includes one or more combinations of pipeline pressure exceeding the limit, fertilizer solution level below the lower limit, sensor offline, actuator feedback abnormality, and communication link abnormality; the safety interlock control includes one or more combinations of load reduction operation, shutting down the fertilizer pump, shutting down the abnormal branch solenoid valve, and stopping the water supply pump.

[0067] In this embodiment, the edge device includes a safety interlock module. This module continuously monitors various status parameters reported by the end-side device, including pipeline pressure, fertilizer solution level, sensor online status, actuator feedback status, and communication link status. When an anomaly is detected, corresponding graded protection actions are executed according to a preset fault level. The first-level fault is pressure... If a single sensor is offline, an audible and visual alarm will be triggered; a level two fault is pressure. or liquid level Implement load reduction operation; Level 3 fault is pressure. or liquid level If a critical actuator reports an abnormality, it will perform protective actions such as shutting down the fertilizer pump, closing the abnormal branch solenoid valve, and stopping the water supply pump, with a response time of ≤100ms.

[0068] If a sensor goes offline or a parameter slightly exceeds its limit (such as pressure exceeding the upper limit by less than 5%), it is considered a Level 1 fault (warning level). The safety interlock module outputs alarm information to the remote operation and maintenance module and the on-site alarm actuator, but does not actively intervene in the control output. The real-time optimization module adjusts itself in the next cycle.

[0069] If the pressure exceeds the upper limit by more than 10% but does not reach the danger level, or the fertilizer solution level is below the lower limit but not exhausted, it is a level two fault (load reduction level). The safety interlock module will automatically perform load reduction operation, such as reducing the water supply pump frequency by 20% and closing some branch solenoid valves to reduce the system load, while triggering an audible and visual alarm.

[0070] If the pipeline pressure exceeds the safety limit, the fertilizer solution level is lower than the protection limit, the critical actuator feedback is abnormal (e.g., the solenoid valve should be open but feedback is closed), or the communication link is completely interrupted, which constitutes a level three fault (protection level), the safety interlock module will sequentially perform the following protection actions: shut down the fertilizer pump (to prevent dry running without liquid), shut down the abnormal branch solenoid valve (to isolate the fault area), stop the water supply pump (to completely shut down the system), and output the highest level alarm information.

[0071] The safety interlock module requires a response time in milliseconds (e.g., triggering a protection action within 50ms after the pressure sensor detects overpressure), independent of cloud decision-making and edge optimization cycles, ensuring that equipment safety and personnel safety are prioritized in extreme situations.

[0072] For example, in a tomato cultivation facility, a blockage in a branch drip irrigation line caused the pipeline pressure to surge to 0.55 MPa (the safety limit is 0.4 MPa). The safety interlock module detected this anomaly within 80ms of the pressure sensor reporting data, classifying it as a level three fault. It immediately shut off the branch's solenoid valve, reduced the water pump frequency to zero, and triggered a buzzer alarm. The entire process was completed within 200ms, without waiting for the real-time optimization module's 5-minute optimization cycle, effectively preventing a pipeline rupture accident.

[0073] This embodiment possesses independent and rapid safety handling capabilities in the face of sudden anomalies through graded protection and millisecond-level response of safety actions.

[0074] During the network outage self-management period, the offline task queue continuously stores the running data according to the preset caching strategy. Each cached data packet contains at least the following fields: acquisition time (millisecond precision), device number, shed number, strategy version number, job batch number, and data content (including sensor data, execution instructions, feedback values, and alarm records).

[0075] The caching strategy employs a circular buffer design with a preset storage capacity (such as data from the last 7 days or 100,000 records). When the storage capacity approaches its limit, a time-priority and batch-priority strategy is used to delete the earliest data packets that have been confirmed to have been successfully retransmitted, making room for the latest data.

[0076] Specifically, edge devices establish connections with the cloud and perform handshake authentication.

[0077] Edge devices upload cached data packets to the cloud one by one in order of collection time from oldest to newest.

[0078] After receiving the data packet, the cloud performs deduplication based on the policy version number and job batch number in the data packet. If a batch of data has already been uploaded (e.g., due to multiple attempts to re-upload during a network outage resulting in duplicates), the cloud automatically discards the duplicate packets and retains only the data that was successfully received the first time.

[0079] The cloud will integrate the retransmitted data into the historical database as samples for subsequent model training.

[0080] The cloud compares the policy version number in the supplementary data with the latest version number. If the versions are inconsistent, a version difference report is generated in the remote operation and maintenance module for the administrator to confirm whether the model needs to be rolled back or recalibrated.

[0081] For example, in a strawberry cultivation facility, the edge network experienced a network outage at 14:23:05 on June 15, 2025, which lasted approximately 4.5 hours until 18:45:30 on the same day. During this period, the offline task queue cached 326 records, including sensor data every 5 minutes, execution parameters for 5 irrigation and fertilization operations, and 1 pressure limit alarm. After network recovery, the edge devices re-uploaded all 326 records in chronological order. Upon receiving the data, the cloud discovered that records 158-163 had been uploaded repeatedly due to the network interruption; after automatic deduplication, only one unique copy was retained. The administrator viewed the complete operational curve during the outage in the data visualization module, confirming that the system's autonomous operation was normal.

[0082] This embodiment ensures the integrity and traceability of data during network outages by using ordered caching and retransmission of offline task queues, cloud deduplication, and version verification.

[0083] As a preferred embodiment, it further includes: generating water and fertilizer control instructions based on the rolling optimized water and fertilizer strategy; the rolling optimized water and fertilizer strategy is a strategy verified by digital twin simulation.

[0084] Please refer to Figure 2 , Figure 2 This is the second flowchart illustrating a cloud-edge-device collaborative intelligent management method for water and fertilizer control of fruits and vegetables provided by the present invention.

[0085] This invention also provides a cloud-edge-device collaborative intelligent management method for water and fertilizer control of fruits and vegetables, applied in the cloud, the method comprising: 201: Generate a global model by aggregating the local water and fertilizer models sent by edge devices in each greenhouse area through federated learning; 202: The model is trained based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; 203: Generate a basic water and fertilizer strategy based on the trained water and fertilizer control model, and send the basic water and fertilizer strategy to the edge device.

[0086] In this embodiment, the cloud includes a federated learning aggregation module to initialize global model parameters. w 0, distributed to edge devices in each shed area; each edge device trains a model based on local data and calculates gradient updates. Homomorphic encryption is used at each edge. After encryption, the data is uploaded to the cloud; the cloud uses the FedAvg algorithm to aggregate and update parameters from each greenhouse area. , in M The total number of shantytowns, m j For the first j Sample size of each shed area m The total sample size is... The learning rate; The cloud will aggregate the global model w t+1 The model is distributed to each edge device; after receiving the global model, each edge device performs transfer learning fine-tuning based on local data; the fine-tuned model parameters are periodically encrypted and uploaded to the cloud to participate in the next round of federated aggregation.

[0087] The cloud platform aggregates historical operational data, environmental data, and crop growth data (such as yield, sugar content, and single fruit weight) from multiple greenhouse areas. This data is grouped by crop type, growth stage, and strategy version, allowing for a horizontal comparison of water and fertilizer application effectiveness across different greenhouse areas. Comparison dimensions include: irrigation amount per unit area, fertilizer application amount per unit area, water and fertilizer utilization rate (output / input ratio), soil EC change trend, and final yield and quality indicators. By analyzing the relationship between different strategy parameters used in different greenhouse areas and actual output, the cloud platform identifies the optimal strategy combination and generates water and fertilizer strategy recommendations.

[0088] For example, in a certain park with eight tomato greenhouses, greenhouse No. 3 used a higher potassium fertilizer ratio (N:P2O5:K2O=10:10:35) and a lower target EC value (1.8mS / cm) during the fruit expansion stage, resulting in better yield and sugar content than other greenhouses. After horizontal comparison in the cloud, this strategy combination was used as a recommended strategy and automatically distributed to the edge devices of other greenhouses.

[0089] After receiving the recommended water and fertilizer strategy from the cloud, the edge device performs a secondary verification based on local conditions. Verification includes: whether the local soil salinity baseline matches the assumptions of the recommended strategy; whether the local device capabilities (e.g., the maximum flow rate of the fertilizer pump) meet the execution requirements of the recommended strategy; and whether the current growth stage of the local crop is consistent with the applicable stage of the recommended strategy. Only when all the above verifications pass will the edge device include the recommended strategy in the next round of optimization calculations. If the verification fails, the edge device returns a verification failure report to the cloud, explaining the reason (e.g., the local soil EC baseline is higher than the applicable range of the recommended strategy). The cloud then reviews the report and decides whether to forcibly issue or adjust the recommended parameters.

[0090] As a preferred embodiment, the method further includes: performing digital twin simulation verification on the rolling optimized water and fertilizer strategy sent by the edge device in the digital twin fruit and vegetable planting environment; the rolling optimized water and fertilizer strategy is a strategy after real-time rolling optimization of the basic water and fertilizer strategy; and sending the simulated and verified water and fertilizer strategy to the edge device.

[0091] In this embodiment, the soil hydrodynamic model (simplified form of the Richards equations) is as follows: , in, Soil moisture content, Unsaturated hydraulic conductivity h For pressure head, S This refers to the root system's water absorption.

[0092] Solute transport model (convection-dispersion equation): , in cThis refers to the solute concentration (EC value). D The hydrodynamic dispersion coefficient, q This refers to the Darcy flow velocity.

[0093] After generating pump and valve execution commands, the edge device first uploads the strategy to be executed to the cloud-based digital twin simulation module. The digital twin simulation module inputs the current soil conditions (moisture content, EC profile distribution) and the weather forecast for the next 6 hours. The digital twin simulation module simulates the soil moisture and EC change curves after the strategy is executed, predicting the dynamic response for the next 6 hours. It then determines whether the prediction results are within the preset safety range (e.g., EC peak value not exceeding 2.8 mS / cm, moisture content not lower than 22% of the wilting point). Only when the simulation results meet all safety conditions is the cloud-based device returned a verification pass command, allowing the strategy to be issued and executed; otherwise, a verification failure and risk warning are returned, and the edge device re-optimizes or reverts to the previous safety strategy.

[0094] After the strategy is actually executed, the end device will transmit the measured data (actual moisture content change, EC change) back to the cloud; the digital twin simulation module will compare the measured data with the simulation prediction results and calculate the deviation; if the deviation exceeds the preset threshold (such as moisture content RMSE>2%), the soil hydraulic parameters (such as saturated hydraulic conductivity Ks, diffusion D) will be automatically corrected; the corrected twin model parameters will be saved for the next simulation verification.

[0095] This embodiment constructs a virtual image of the fruit and vegetable growing environment in the cloud, realizing dual closed-loop control of simulation verification before execution and feedback correction after execution.

[0096] The intelligent water and fertilizer management system for fruits and vegetables based on cloud-edge-device collaboration provided by the present invention will be described below. The intelligent water and fertilizer management system for fruits and vegetables based on cloud-edge-device collaboration described below can be referred to in correspondence with the intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration described above.

[0097] This invention also provides a cloud-edge-device collaborative intelligent water and fertilizer management system for fruits and vegetables, applied to edge devices. The system includes: a first module for determining the state vector of the planting site based on multi-source detection data from the fruit and vegetable planting site; a second module for performing real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector and sending water and fertilizer control instructions to the edge device; the basic water and fertilizer strategy is generated based on a water and fertilizer control model trained in the cloud; a third module for receiving execution status feedback data sent by the edge device; and a fourth module for performing autonomous water and fertilizer management of fruits and vegetables in the event of a network outage between the edge device and the cloud, based on the status feedback data and the cached water and fertilizer control model.

[0098] This invention also provides a cloud-edge-device collaborative intelligent water and fertilizer management system for fruits and vegetables, applied in the cloud. The system includes: a fifth module, used to aggregate local water and fertilizer models sent by edge devices in each greenhouse area through federated learning to generate a global model; a sixth module, used to train the model based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; and a seventh module, used to generate a basic water and fertilizer strategy based on the trained water and fertilizer control model and send the basic water and fertilizer strategy to the edge devices.

[0099] In this embodiment, the cloud-edge-device collaborative control process for the entire growth cycle of melon cultivation is as follows: The end-side equipment continuously collects data on soil moisture content, soil EC value, soil temperature, air temperature and humidity, light intensity, fertilizer solution EC value, fertilizer solution pH value, pipeline flow rate, pipeline pressure, and fertilizer solution level at the planting site. At the same time, it collects the execution feedback status (such as operating frequency, opening and closing status, and fault codes) of the water supply pump, fertilizer pump, and solenoid valve.

[0100] The data cleaning and fusion module of the edge device performs timestamp alignment, outlier removal (such as removing outliers with negative soil moisture content), missing value filling (such as taking the average of the time before and after when a single data point of a soil moisture sensor is missing), and multi-point spatial fusion (such as the weighted average of four soil moisture sensors) on the raw data uploaded by the end device to form a standardized planting site state vector.

[0101] Based on the past three years of planting data from the park, the current water and fertilizer requirements of melon varieties, and the weather forecast data for this season, the cloud platform trains or updates the water and fertilizer control model. The cloud platform then distributes the basic water and fertilizer strategies (including target moisture content ranges, target EC values, nitrogen, phosphorus, and potassium ratios, and pulse parameters for the seedling stage, vine extension stage, flowering and fruit setting stage, fruit expansion stage, and maturity stage), model parameters, and version number (e.g., V2.3.1) to the edge devices.

[0102] Edge devices use cloud-based basic strategies as initial values, and the current field state vector includes root zone moisture content (measured 23%, target lower limit 25%), fertilizer solution EC value (measured 1.5 mS / cm, target 1.8 mS / cm), soil EC value (measured 2.1 mS / cm, safety threshold 2.5 mS / cm), and flow rate (measured 2.1 m³ / cm). 3 / h, target 2.5m 3The optimal control parameters for this round are calculated using a water and fertilizer execution process prediction model with accumulated temperature correction, with feedback parameters including water / h and pressure (measured at 0.32 MPa, safety limit 0.40 MPa). Before generating water and fertilizer control commands, the edge device uploads the water and fertilizer strategy to be executed to the cloud-based digital twin simulation module for simulation verification. The digital twin simulation module simulates soil moisture and EC changes over the next 6 hours. After confirming that the EC peak does not exceed 2.8 mS / cm and the moisture content is not lower than 22% of the wilting point, the verification is considered successful. The edge device then sends the water supply pump frequency (should be adjusted from 42Hz to 48Hz), fertilizer pump speed (should be adjusted from 1200 rpm to 1450 rpm), and solenoid valve opening and closing durations (100 seconds for pre-wetting, 60 seconds x 2 times for fertilization, and 80 seconds for flushing) to the end-side device.

[0103] The end-side equipment executes pump and valve commands. The water supply pump operates at 48Hz, the fertilizer pump injects fertilizer solution at 1450rpm, and each solenoid valve opens and closes sequentially. During execution, the end-side equipment returns real-time data on fertilizer solution EC value, flow rate, pressure, and execution status (such as pump operating current and valve opening / closing signals) to the edge equipment. After execution, the measured data is transmitted back to the cloud, and the digital twin simulation module compares the measured data with the simulation predictions. If the deviation is within the allowable range, the digital twin model parameters remain unchanged.

[0104] During the mid-fruit expansion stage of the melons, the park's network was interrupted due to a fiber optic cable being severed. After failing to receive a heartbeat response from the cloud three times consecutively, the edge devices automatically entered a network-off autonomous mode. The current fruit expansion stage model V2.3.1 in the model cache module was still valid (valid until July 31, 2025), and the edge devices continued to use this model for local rolling optimization. From the moment the network was disconnected, the offline task queue cached all collected data, optimization calculation results, and execution feedback records at a frequency of one data packet every 5 minutes. Three days later, when the network was restored, the edge devices retransmitted all cached data (a total of 864 data packets) in chronological order. After receiving the data, the cloud performed a consistency check based on the version number, confirming that V2.3.1 was consistent with the version before the network outage, and added the retransmitted data to the historical database for the next round of federated learning training.

[0105] This embodiment verifies the system's ability to operate continuously and ensure data integrity under abnormal network conditions by going through the entire process from data acquisition, cleaning and fusion, model distribution, rolling optimization, execution feedback to network outage self-governance and retransmission.

[0106] The intelligent water and fertilizer management method and system for fruits and vegetables based on cloud-edge-device collaboration provided by this invention can be widely applied to integrated water and fertilizer management scenarios involving multiple greenhouses, crops, and equipment in facility agriculture. This invention effectively solves the problems of strong cloud dependency, inability to operate continuously after network outages, and difficulty in integrating heterogeneous device protocols in traditional systems through a three-layer cloud-edge-device architecture with division of labor and collaboration. This improves the system's stability, scalability, and maintainability, and has good industrial practical value and promising prospects for widespread application.

[0107] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 301, a communications interface 302, a memory 303, and a communications bus 304, wherein the processor 301, the communications interface 302, and the memory 303 communicate with each other through the communications bus 304. The processor 301 can call logical instructions in the memory 303 to execute a cloud-edge-device collaborative intelligent water and fertilizer management method for fruits and vegetables, applied to edge devices. This method includes: determining the state vector of the planting site based on multi-source detection data from the fruit and vegetable planting site; performing real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector, and sending water and fertilizer control instructions to the edge device; the basic water and fertilizer strategy is generated based on a water and fertilizer control model trained in the cloud; receiving execution status feedback data sent by the edge device; and performing autonomous water and fertilizer management of fruits and vegetables in the event of a network outage between the edge device and the cloud, based on the status feedback data and the cached water and fertilizer control model; or, executing a cloud-edge-device collaborative intelligent water and fertilizer management method for fruits and vegetables, applied to the cloud. This method includes: aggregating local water and fertilizer models sent by edge devices in each greenhouse area through federated learning to generate a global model; training the model based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; generating a basic water and fertilizer strategy based on the trained water and fertilizer control model, and sending the basic water and fertilizer strategy to the edge device.

[0108] Furthermore, the logical instructions in the aforementioned memory 303 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the cloud-edge-device collaborative intelligent water and fertilizer management method for fruits and vegetables provided by the above methods, applied to edge devices. The method includes: determining the state vector of the planting site based on multi-source detection data of the fruit and vegetable planting site; performing real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector, and sending water and fertilizer control instructions to the edge device; the basic water and fertilizer strategy is based on a water and fertilizer control model trained in the cloud. The model is generated; the execution status feedback data sent by the receiving end-side device is received; in the case of network disconnection between the edge device and the cloud, autonomous water and fertilizer management of fruits and vegetables is performed based on the status feedback data and the cached water and fertilizer control model; or, a cloud-edge-device collaborative intelligent water and fertilizer management method for fruits and vegetables is executed and applied to the cloud. This method includes: aggregating the local water and fertilizer models sent by the edge devices of each greenhouse area through federated learning to generate a global model; training the model based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; generating a basic water and fertilizer strategy based on the trained water and fertilizer control model, and sending the basic water and fertilizer strategy to the edge device.

[0110] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration provided by the above methods, applied to edge devices. The method includes: determining the state vector of the planting site based on multi-source detection data of the fruit and vegetable planting site; performing real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector, and sending water and fertilizer control instructions to the edge device; the basic water and fertilizer strategy is generated based on a water and fertilizer control model trained in the cloud; receiving execution status feedback data sent by the edge device; performing autonomous water and fertilizer management of fruits and vegetables in the event of a network outage between the edge device and the cloud, based on the status feedback data and the cached water and fertilizer control model; or, executing the intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration, applied to the cloud. The method includes: aggregating local water and fertilizer models sent by edge devices in each greenhouse area through federated learning to generate a global model; training the model based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model; generating a basic water and fertilizer strategy based on the trained water and fertilizer control model, and sending the basic water and fertilizer strategy to the edge device.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent management and control of water and fertilizer in fruits and vegetables based on cloud-edge-device collaboration, characterized in that, Applied to edge devices, the method includes: Based on multi-source detection data from fruit and vegetable planting sites, determine the state vector of the planting site; Based on the planting site state vector, the basic water and fertilizer strategy is continuously optimized in real time, and water and fertilizer control commands are sent to the end device; the basic water and fertilizer strategy is generated based on the water and fertilizer control model trained in the cloud. Receive execution status feedback data sent by the terminal device; In the event of a network outage between the edge device and the cloud, autonomous management of water and fertilizer for fruits and vegetables is achieved based on the status feedback data and the cached water and fertilizer control model.

2. The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration according to claim 1, characterized in that, The process of determining the state vector of the planting site based on multi-source detection data from the fruit and vegetable planting site includes: Protocol adaptation and data cleaning and fusion are performed on the multi-source detection data of the fruit and vegetable planting site to obtain the planting site state vector; the fields of the planting site state vector include one or more combinations of timestamp, greenhouse number, soil moisture content, soil conductivity, soil temperature, air temperature, air relative humidity, light intensity, fertilizer solution conductivity, fertilizer solution pH, pipeline flow rate, pipeline pressure and fertilizer solution level.

3. The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration according to claim 1, characterized in that, The step of performing real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector and sending water and fertilizer control commands to the end-side equipment includes: The planting site state vector and the target irrigation amount and target fertilizer amount in the basic water and fertilizer strategy are input into the water and fertilizer execution process prediction model to obtain the pump valve control quantity output by the water and fertilizer execution process prediction model, and the water and fertilizer control command is sent to the end-side equipment based on the pump valve control quantity; the water and fertilizer execution process prediction model takes the minimum deviation between the predicted planting state and the target planting state as the optimization objective, and updates the effective accumulated temperature during the water and fertilizer execution process, with constraints including root zone salt accumulation constraints.

4. The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration according to claim 1, characterized in that, The offline autonomous management of water and fertilizer control for fruits and vegetables based on the state feedback data and the cached water and fertilizer control model includes: If the cached water and fertilizer control model and corresponding water and fertilizer strategy are within their validity period, the cached water and fertilizer control model and the state feedback data are used for rolling optimization of the water and fertilizer strategy and water and fertilizer control; or, If the cached water and fertilizer control model is valid and the water and fertilizer strategy expires, based on the state feedback data and the cached water and fertilizer control model, the parameters of the water and fertilizer strategy will be reverted to the preset target water and fertilizer control value; or, When the cached water and fertilizer control model has exceeded the validity period of the water and fertilizer strategy, the water and fertilizer strategy is controlled to enter the nutrient stress buffer control mode; the nutrient stress buffer control mode is used to adjust the water and fertilizer strategy according to the preset load reduction rate.

5. The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration according to claim 1, characterized in that, Also includes: The status of the fruit and vegetable planting site is determined based on the execution status feedback data. In the event of abnormal conditions at the fruit and vegetable planting site, safety interlock control shall be executed. Abnormal conditions at the fruit and vegetable planting site include one or more of the following combinations: pipeline pressure exceeding the limit, fertilizer solution level below the lower limit, sensor offline, actuator feedback abnormality, and communication link abnormality. The safety interlock control includes one or more of the following combinations: load reduction operation, shutting down the fertilizer pump, shutting down the abnormal branch solenoid valve, and stopping the water supply pump.

6. The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration according to any one of claims 1 to 5, characterized in that, Also includes: The water and fertilizer control instructions are generated based on the rolling optimized water and fertilizer strategy; The optimized water and fertilizer strategy is a strategy verified through digital twin simulation.

7. A method for intelligent management and control of water and fertilizer in fruits and vegetables based on cloud-edge-device collaboration, characterized in that, Applied to the cloud, the method includes: A global model is generated by aggregating the local water and fertilizer models sent by edge devices in each greenhouse area through federated learning. The model is trained based on the global model and the water and fertilizer control training set to obtain a trained water and fertilizer control model. A basic water and fertilizer strategy is generated based on the trained water and fertilizer control model, and the basic water and fertilizer strategy is sent to the edge device.

8. The intelligent water and fertilizer management method for fruits and vegetables based on cloud-edge-device collaboration according to claim 7, characterized in that, Also includes: In a digital twin-based fruit and vegetable growing environment, the optimized water and fertilizer strategy sent by edge devices is verified through digital twin simulation. The rolling optimized water and fertilizer strategy is a strategy that is optimized in real time from the basic water and fertilizer strategy. Send the simulated and verified water and fertilizer strategy to the edge device.

9. A cloud-edge-device collaborative intelligent water and fertilizer management system for fruits and vegetables, characterized in that, The system, applied to edge devices, includes: The first module is used to determine the state vector of the planting site based on multi-source detection data from the fruit and vegetable planting site. The second module is used to perform real-time rolling optimization of the basic water and fertilizer strategy based on the planting site state vector, and send water and fertilizer control commands to the end device; the basic water and fertilizer strategy is generated based on the water and fertilizer control model trained in the cloud. The third module is used to receive execution status feedback data sent by the end-side device; The fourth module is used to perform autonomous management of water and fertilizer for fruits and vegetables in the event of a network outage between the edge device and the cloud, based on the status feedback data and the cached water and fertilizer control model.

10. A cloud-edge-device collaborative intelligent water and fertilizer management system for fruits and vegetables, characterized in that, The system, applied in the cloud, includes: The fifth module is used to aggregate the local water and fertilizer models sent by the edge devices of each greenhouse area through federated learning to generate a global model; The sixth module is used to train the model based on the global model and the water and fertilizer control training set to obtain the trained water and fertilizer control model. The seventh module is used to generate a basic water and fertilizer strategy based on the trained water and fertilizer control model, and send the basic water and fertilizer strategy to the edge device.

Images