A method and system for constructing a smart farm

Abstract

This invention provides a method and system for constructing a smart farm, belonging to the field of smart agriculture technology. The method includes: simultaneously collecting multi-source environmental data such as soil moisture content, soil electrical conductivity, and canopy temperature and humidity within the farm area; calculating a water suitability index, a nutrient balance index, and an environmental coordination index based on the data, and using the environmental coordination index to correct the water and nutrient indices; generating a comprehensive environmental state index through a fusion model; inputting the index into a neural network decision model optimized by an improved particle swarm optimization algorithm, outputting growth regulation priorities represented by a probability distribution; generating specific farmland collaborative management decision instructions based on these priorities, and controlling intelligent agricultural machinery to execute them. This invention achieves coupled evaluation and intelligent priority decision-making for multiple environmental factors such as water, fertilizer, air, and heat, enabling precise identification and regulation of the most critical limiting factors, thereby improving resource utilization efficiency and management precision in agricultural production.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to a method and system for constructing a smart farm. Background Technology

[0002] The field of smart agriculture has widely applied IoT sensor technology to monitor farmland environments and to achieve automated control of single factors such as irrigation and fertilization based on threshold rules. However, existing technologies still face some systemic challenges in practical applications.

[0003] First, while existing systems can collect multi-dimensional data such as soil moisture, nutrients, and weather conditions, they typically process each environmental factor independently. Decisions are often triggered by data from a single sensor, such as deciding on irrigation based solely on soil moisture, failing to fully consider the real-time impact of weather conditions like temperature and humidity on crop water requirements, and lacking quantitative assessment methods for the interactions and combined effects of factors such as water, fertilizer, air, and heat. This leads to fragmented management strategies, making it difficult to coordinate responses to complex and ever-changing field environments.

[0004] Secondly, most intelligent models currently used in agricultural decision-making employ general predictive or classification models. These models typically process raw or simply preprocessed data directly, and their structures are not designed for the specific task of multi-factor coupling analysis and prioritization decision-making. The model outputs are often multiple parallel and potentially independent control recommendations, failing to clearly indicate the core limiting factors that most urgently require intervention under the current comprehensive conditions, thus posing difficulties for precise implementation.

[0005] Finally, most systems operate on fixed rules or model parameters, lacking a closed-loop learning mechanism that continuously optimizes the core computational and decision-making models based on feedback from execution results. Consequently, these systems struggle to adapt to the dynamic changes in crop growth stages and the long-term evolution of the environment, and their long-term accuracy and adaptability need improvement.

[0006] Therefore, there is a need for a new smart farm management method that can achieve comprehensive quantification of the state of multiple environmental factors and make intelligent priority decisions and continuous self-optimization based on this. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for constructing a smart farm, in order to solve the problems of fragmented environmental factor management, lack of priority output in decision-making models, and insufficient system adaptability in the prior art.

[0008] To achieve the above objectives, the present invention provides a method for constructing a smart farm, comprising the following steps: Step S1: Synchronously collect multi-source environmental data within the farm area; Step S2: Based on the collected multi-source environmental data, the water suitability index, nutrient balance index, and environmental coordination index are calculated respectively using a preset calculation model. The water suitability index and nutrient balance index are then corrected using a correction model based on the environmental coordination index to obtain the corrected water suitability index and the corrected nutrient balance index. Finally, the corrected water suitability index, the corrected nutrient balance index, and the environmental coordination index are combined using a fusion model to generate a comprehensive environmental status index. Step S3: Input the comprehensive environmental state index and its sub-indices into the environmental factor coupled decision model, and output the growth regulation priority; Step S4: Generate specific farmland collaborative management decision instructions based on the control direction and intensity indicated by the growth control priority. Step S5: Control the intelligent agricultural machinery and equipment to execute farmland collaborative management decision commands.

[0009] The present invention also provides a smart farm construction system for performing the smart farm construction method described above, comprising: The multi-source data acquisition module is used to simultaneously collect multi-source environmental data within the farm area, including soil volumetric water content, soil electrical conductivity, canopy air temperature, and canopy air relative humidity. The environmental index calculation module is used to calculate the water suitability index, nutrient balance index, and environmental coordination index based on multi-source environmental data and through a preset calculation model. The environmental coordination index is used to correct the water suitability index and nutrient balance index through a correction model to obtain the corrected water suitability index and the corrected nutrient balance index. Finally, the corrected water suitability index, the corrected nutrient balance index, and the environmental coordination index are combined through a fusion model to generate a comprehensive environmental status index. The growth regulation decision module is used to input the comprehensive environmental state index and its sub-indices into the environmental factor coupled decision model and output the growth regulation priority; the environmental factor coupled decision model is a dedicated model based on an improved particle swarm optimization algorithm to optimize neural network parameters; The collaborative management decision-making module is used to generate specific farmland collaborative management decision-making instructions based on the control direction and intensity indicated by the growth control priority; and The intelligent execution control module is used to control intelligent agricultural machinery and equipment to execute farmland collaborative management decision-making instructions.

[0010] Therefore, the present invention employs the above-described method and system for constructing a smart farm, and the beneficial technical effects are as follows: (1) This invention generates independent indices for water, nutrients, and atmospheric and thermal environments through a preset calculation model, and introduces an environmental coordination index as a coupling factor. The water and nutrient indices are dynamically adjusted through a correction model. Finally, a unified comprehensive environmental state index is generated using a fusion model. This technical approach deeply integrates and quantitatively correlates the originally isolated water, fertilizer, air, and heat data, enabling the system to assess the comprehensive state under the interaction of multiple environmental factors. This provides a precise and quantitative basis for integrated collaborative management, overcoming the drawbacks of independent processing of each factor and fragmented decision-making in existing technologies.

[0011] (2) This invention constructs a feedforward neural network based on an improved particle swarm optimization algorithm as an environmental factor coupled decision-making model. This model is specifically designed for multi-factor coupled analysis and priority decision-making tasks. Its output is not a parallel suggestion, but a probability distribution vector normalized by the Softmax function, representing the priority of water, nutrient, and atmospheric heat regulation. This output can clearly indicate the core limiting factors that most urgently need to be regulated, solving the problem of ambiguous output of general models and difficulty in directly guiding precise execution.

[0012] (3) In optimizing the decision model parameters, the fitness function of the improved particle swarm optimization algorithm in this invention not only considers decision accuracy (deviation from the historical best decision), but also introduces estimated resource consumption as one of the optimization objectives, and adjusts it through weight balancing. This makes the model tend to choose decision strategies with lower resource consumption when achieving similar control effects during the training process. This design embeds economy and resource efficiency into the decision logic, enabling the system to adapt to different growth stages of crops and environmental changes. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a method for constructing a smart farm according to the present invention; Figure 2 A logic diagram for the integration of environmental index calculation and correction; Figure 3 A flowchart for the construction and reasoning process of a decision-making model coupled with environmental factors. Detailed Implementation

[0014] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0016] Example 1 Reference Figures 1-3This embodiment provides a method for constructing a smart farm, applied to a multi-span glass greenhouse for growing tomatoes. This method, through the deployment of a sensor network and intelligent execution equipment, achieves coordinated and precise management of the water, fertilizer, air, and heat environments in different areas (micro-plots) within the farm.

[0017] Step S1: Synchronously collect multi-source environmental data within the farm area.

[0018] In this embodiment, the following sensor network is deployed to collect data synchronously: Soil volumetric water content ( A tubular soil moisture meter using the FDR (Frequency Domain Reflectometry) principle was used in the greenhouse ridge planting area at 10-meter intervals. A 10-meter grid was laid out in different micro-plots, at a depth of the main root layer of the crop, and data was collected regularly.

[0019] Soil electrical conductivity ( ): Synchronous data acquisition is achieved using an EC sensor integrated with a moisture sensor.

[0020] Canopy air temperature ( ) and canopy air relative humidity ( ): Digital temperature and humidity sensors with protective covers are installed near the canopy of each micro-area plot and collected data periodically.

[0021] All sensor data is transmitted to the IoT gateway via a wireless network and timestamped to form a synchronous multi-source environmental data packet covering each micro-domain.

[0022] Step S2: Based on the collected multi-source environmental data, the water suitability index, nutrient balance index, and environmental coordination index are calculated respectively using a preset calculation model. The water suitability index and nutrient balance index are then corrected using a correction model based on the environmental coordination index to obtain the corrected water suitability index and the corrected nutrient balance index. Finally, the corrected water suitability index, the corrected nutrient balance index, and the environmental coordination index are comprehensively calculated using a fusion model to generate the comprehensive environmental status index.

[0023] This embodiment takes the tomato fruit enlargement period as an example, with the following preset parameters: Calculate model parameters: Optimal soil moisture content (Volume water content), Effective water content range .

[0024] Target conductivity mS / cm, scale parameter .

[0025] Optimal saturated vapor pressure difference kPa, effective control range kPa.

[0026] Correct model parameters: Water stress correction intensity coefficient .

[0027] Nutrient availability correction strength coefficient .

[0028] Environmental coordination critical threshold .

[0029] Fusion model parameters (preset for the reproductive growth stage of tomatoes): Weighting coefficients: , , .

[0030] Suppose a set of data is collected at a certain moment: , mS / cm ℃, .

[0031] Moisture suitability index The calculation formula is: ; in, This represents the optimal soil moisture content for the current growth stage of the crop. The permissible effective water content range; Nutrient balance index The calculation formula is: ; in, The target conductivity for the current crop growth stage. This is a scale parameter characterizing tolerance. Environmental Coordination Index The calculation formula is: ; ; in, Based on the collected canopy air temperature and relative humidity The calculated actual saturated water vapor pressure difference For the optimal saturated water vapor pressure difference, To effectively regulate the scope.

[0032] The modified model is as follows: Corrected Moisture Suitability Index ,in, This is the water stress correction intensity coefficient, with a value ranging from 0.1 to 0.5. This represents the critical threshold for environmental coordination. Corrected nutrient balance index ,in, The intensity coefficient for nutrient availability is adjusted, with a value ranging from 0.05 to 0.3.

[0033] The modified operation of this invention aims to couple the real-time effects of meteorological environment (air, heat) to the assessment of soil water and fertilizer status. When the environmental compatibility index is low (e.g., high temperature, low humidity), crop transpiration is aggravated, which may lead to increased actual water stress under the same soil moisture conditions and may affect nutrient availability. Therefore, utilizing... right and The revisions can make the assessment more closely reflect the physiological sensations of crops in real-world complex environments.

[0034] The fusion model is specifically a weighted geometric mean function and an environmental comprehensive state index. The calculation formula is: ; Among them, the weighting coefficient , , This is a fixed value preset according to the crop's vegetative or reproductive growth stage, and it meets the following requirements: ; Based on the above settings, the comprehensive environmental state index in this embodiment is approximately 0.741.

[0035] Step S3: Input the comprehensive environmental state index and its sub-indices into the environmental factor coupling decision model (the environmental factor coupling decision model is a special model based on the improved particle swarm algorithm to optimize the neural network parameters), and output the growth regulation priority.

[0036] In this embodiment, the environmental factor coupled decision model is a specific feedforward neural network with the following structure: 5 nodes in the input layer (corresponding to...) , , , Fertility period coding The network consists of a hidden layer with 8 neurons (using the ReLU activation function) and an output layer with 3 nodes (corresponding to the initial scores for water, nutrient, and heat regulation). The parameters (weights and biases) of the network are determined by optimization using the improved particle swarm optimization algorithm described below.

[0037] a) Model structure definition: Construct a feedforward neural network as the main decision network, whose input layer nodes correspond to input features, including the comprehensive environmental state index. Corrected Moisture Suitability Index Corrected nutrient balance index Environmental Coordination Index and crop growth period coding The output layer nodes correspond to the growth regulation priority. The dimension; the number of hidden layers is 1 to 3; Growth regulation priority The output process is as follows: The output layer of the main decision network consists of three nodes, each outputting the initial score for water regulation. Initial score of nutrient regulation Initial score of thermal environment regulation In this embodiment, the values ​​are 1.2, 0.8, and 0.5, respectively. Subsequently, the three initial scores are normalized into a probability distribution using the Softmax function, calculated as follows: ; in, ; Growth regulation priority That is, the probability distribution vector , The values ​​were 0.46, 0.31, and 0.25, respectively. The category corresponding to the maximum value was determined to be the most urgent limiting factor to be regulated. Therefore, the system determined that the most urgent limiting factor to be regulated was moisture.

[0038] The system performs this inference process independently for each micro-domain, resulting in distinct GP vectors.

[0039] b) Parameter optimization: An improved particle swarm optimization algorithm is used to globally optimize the connection weights and bias parameters of the main decision network; the improvement is reflected in the fitness function design, the formula of which is: ; in, Encoding the historically optimal control decision, For the network's predicted output, To execute based on historical data statistics The estimated resource consumption required for the corresponding decision-making and To balance the weights, , .

[0040] c) Algorithm execution: The particle swarm size is set to 50 to 200, and the particle velocity update formula incorporates an inertial weight that is dynamically adjusted based on fitness ranking. , The initial value is 0.9, and the final value at the end of the iteration is 0.4, which is the acceleration constant. .

[0041] During training, the improved particle swarm optimization algorithm employs inertial weights that are dynamically adjusted based on fitness ranking. After each iteration, particles are ranked according to their fitness, and the top 20% of particles are given an inertia weight. Values ​​are taken in the range of 0.3 to 0.5 to promote local development; for the bottom 50% of particles, their inertia weight is adjusted. Values ​​are taken in the range of 0.8 to 1.0 to encourage global exploration, thereby optimizing neural network parameters more effectively.

[0042] d) Model inference: Initialize the main decision network using the optimized parameters, feed real-time input features into the network, and calculate the growth regulation priority through forward propagation. .

[0043] Step S4: Generate specific farmland collaborative management decision instructions based on the control direction and intensity indicated by the growth control priority.

[0044] Generate specific farmland collaborative management decision instructions, and perform the following operations: Step S41: Determine the priority of growth regulation The category of target regulatory factors corresponding to the maximum probability in the middle In this embodiment, the substance is water.

[0045] Step S42: Based on the target regulatory factor category Read the corresponding real-time sensor data variables from the preset control parameter mapping table. With target threshold The mapping relationship is: if If it is water, then , t If it is nutrients, then , If it is a hot environment, then , t .

[0046] Step S43: Calculate the difference And through the corresponding linear control function in the mapping table Calculate the specific equipment control parameter values. ;in This is the proportionality coefficient. As the baseline value, The specific control parameters output vary depending on the device. These represent irrigation duration, fertilizer pump speed, target fan speed, or target shade net opening, respectively.

[0047] Based on the above parameter settings, calculate The duration is 70 seconds. The system generates the instruction "Start the drip irrigation system in the coordinate area (X, Y) for 70 seconds of irrigation".

[0048] At the same time, the system supports generating differentiated fertilization strategies: for example, if another plot of land... Showing the highest nutrient priority and Under suitable conditions, the system may generate an instruction to "apply high-potassium compound fertilizer to this area at a rate of 5 kg / acre." Display of atmospheric and thermal environment ( If a relevant parameter is prioritized for control, it may generate instructions to "turn on the circulating fan" or "start the shade net." This allows for differentiated fertilization, pesticide application, and environmental control strategies for different micro-plots.

[0049] Step S5: Control the intelligent agricultural machinery and equipment to execute farmland collaborative management decision commands.

[0050] Command issuance: Specific commands generated by the collaborative management decision-making module (e.g., drip irrigation in area A for 70 seconds, simultaneous fertilization until...). The mS / cm) is sent to the intelligent execution control module. This module has an embedded device protocol converter that encapsulates general instructions into corresponding standard industrial protocol data packets based on the target execution terminal's model and communication interface, for example: For intelligent water and fertilizer integrated machines that support Modbus TCP, the message is encapsulated as function code 03 (read holding register) and 06 (write single register), specifying the pump start / stop register address, inverter frequency register address, and fertilizer pump speed register address.

[0051] For unmanned agricultural drones that support the CAN bus protocol, the CAN data frame is encapsulated as a specific identifier (ID) and includes the boundary coordinates of the work area, the nozzle switch command, and the flight speed parameters.

[0052] For IoT environmental control devices based on 4G / 5G (such as fans and film reel rollers), structured JSON commands are published to the corresponding topic of the device via the MQTT protocol.

[0053] Command Transmission and Response: The encapsulated command data packet is transmitted via wired industrial Ethernet, wireless LoRa gateway, or public cellular network within the farm to the embedded controller or IoT terminal of the target intelligent agricultural machinery. Upon receiving the command, the device controller first verifies and parses it, and then returns a command reception confirmation signal to the system.

[0054] Operation Execution and Monitoring: The intelligent equipment begins its operation. The system monitors the operation process in real time using real-time status data transmitted back by the equipment (such as the instantaneous flow rate of the irrigation system, the real-time EC value of the fertilizer applicator, and the current waypoint and remaining amount of pesticide in the plant protection machine). The system presets tolerance ranges for key parameters (such as flow rate deviation ±5%). Once the monitored data exceeds the limit, an alarm is immediately triggered, and pause or adjustment commands can be executed according to preset strategies.

[0055] Execution Result Feedback: Upon completion of the operation, the execution terminal uploads an execution summary report, which includes: final execution time, actual total water / fertilizer consumption, coverage area, actual average EC value, and start and end timestamps. This report is bound and stored with the original decision instructions generated in step S4, forming a traceable decision-execution record.

[0056] Example 2 This embodiment provides a method for constructing a smart farm, applicable to large-scale cornfield planting areas.

[0057] Step S1: Multi-source environmental data acquisition.

[0058] Micro-plots were divided into 50m × 50m grids within the cornfield. Soil sensors were deployed at the depth of the crop's main root system in each plot to measure soil volumetric water content and soil electrical conductivity. Simultaneously, temperature and humidity sensors were deployed near the canopy to measure canopy air temperature and relative humidity. All sensors collected data every 15 minutes, which was then aggregated via a wireless gateway deployed in the field and timestamped to create a synchronous multi-source environmental data packet covering all micro-plots.

[0059] Step S2: Environmental index calculation and correction.

[0060] Based on the physiological needs of maize at different growth stages, various parameters in the calculation model, intensity coefficients in the correction model, and weight allocations in the fusion model are pre-set. The entire growth period of maize is divided into four main stages: seedling stage, jointing stage, tasseling and silking stage, and grain-filling stage.

[0061] Seedling stage (0-20 days after sowing): The optimal soil moisture content is set at 65% of field capacity (approximately 22% volumetric water content), with an effective moisture content range of ±5%. The target electrical conductivity is set at 0.8 mS / cm, and the scale parameter is 0.2. The optimal saturated vapor pressure difference is set at 1.2 kPa, with an effective control range of ±0.6 kPa.

[0062] During the jointing stage (21-50 days after sowing): the optimum soil moisture content is increased to 75% of field capacity (approximately 25% volumetric water content), and the effective moisture content range is narrowed to ±4%. The target electrical conductivity is increased to 1.2 mS / cm, while the scale parameter remains at 0.2. The optimum saturated vapor pressure difference is set at 1.5 kPa, and the effective control range is set at ±0.8 kPa.

[0063] During the tasseling and silking stage (51–70 days after sowing): the optimum soil moisture content is further increased to 80% of field capacity (approximately 27% volumetric water content), and the effective moisture content range is further narrowed to ±3%. The target electrical conductivity is set at 1.5 mS / cm, and the scale parameter is 0.2. The optimum saturated vapor pressure difference is set at 1.8 kPa, and the effective control range is set at ±0.7 kPa.

[0064] During the grain-filling stage (71–100 days after sowing): the optimum soil moisture content is adjusted back to 70% of field capacity (approximately 24% volumetric water content), and the effective moisture content range is widened to ±4%. The target electrical conductivity is set at 1.2 mS / cm, and the scale parameter is 0.2. The optimum saturated vapor pressure difference is set at 1.6 kPa, and the effective control range is set at ±0.7 kPa.

[0065] Taking data collected at a certain moment in a micro-plot during the jointing stage as an example: the measured soil volumetric water content was 23%, lower than the optimal value of 25% for this stage; the measured soil electrical conductivity was 1.1 mS / cm, slightly lower than the target value of 1.2 mS / cm; the canopy air temperature was 28℃, and the canopy air relative humidity was 55%. Based on the temperature and humidity values, the actual saturated vapor pressure difference was calculated to be approximately 1.8 kPa using the Magnus formula.

[0066] Based on the preset calculation model, the following sub-indices were calculated: the water suitability index was 0.5, the nutrient balance index was approximately 0.882, and the environmental coordination index was 0.625.

[0067] The critical threshold for environmental coordination is set at 0.6. The current environmental coordination index is 0.625, which is greater than the critical threshold, indicating that meteorological conditions are acceptable and no correction is needed. If the environmental coordination index falls below the critical threshold, the water suitability index and nutrient balance index will be corrected according to the correction model, with correction intensity coefficients of 0.3 and 0.2, respectively.

[0068] The fusion model employs a weighted geometric mean function, with the weights of the water suitability index, nutrient balance index, and environmental coordination index at the jointing stage preset to 0.5, 0.3, and 0.2, respectively. The calculated comprehensive environmental state index is approximately 0.617.

[0069] Step S3: Output growth regulation priority.

[0070] The comprehensive environmental state index, water suitability index, nutrient balance index, environmental coordination index, and crop growth stage code (jointing stage code is 2) are input into the environmental factor coupled decision model. This model is a feedforward neural network optimized with an improved particle swarm optimization algorithm, with 5 nodes in the input layer, 8 nodes in the hidden layer, and 3 nodes in the output layer, corresponding to the initial scores for water regulation, nutrient regulation, and atmospheric and thermal environment regulation, respectively. The three initial scores are normalized to a probability distribution vector using the Softmax function, for example, obtaining [0.65, 0.20, 0.15], indicating that the most urgent limiting factor for the current micro-domain is water.

[0071] Step S4: Generate specific farmland collaborative management decision instructions.

[0072] First, determine the category of the target regulatory factor corresponding to the highest probability in the growth regulation priority; in this case, it is water. Read the corresponding real-time sensor data variables and target thresholds from a pre-set regulation parameter mapping table. For water regulation, the real-time variable is soil volumetric water content, and the target threshold is set to the optimal soil water content of the current growth stage, 25%. The difference between the measured value and the target threshold is 2 percentage points.

[0073] Based on the preset linear relationship, the irrigation duration is directly proportional to the difference in soil volumetric moisture content, with a proportionality coefficient of 30 min / percentage point. Adding a baseline duration of 5 min, the calculated irrigation duration is 65 min. The system generates the specific instruction: "Start the drip irrigation system in micro-domain A (coordinates X, Y), with an irrigation duration of 65 min."

[0074] Simultaneously, the system independently executes the above inference process for other micro-plots. If the growth regulation priority of another plot shows the highest nutrient priority and a low nutrient balance index, the system will read the real-time variable corresponding to nutrient regulation from the mapping table as soil electrical conductivity, the target threshold as target electrical conductivity, calculate the difference, and generate instructions according to a preset linear relationship: the fertilizer pump speed is proportional to the difference in soil electrical conductivity, with a proportionality coefficient of 50 (L / h) / (mS·cm). -1 The system generates commands such as "Apply water-soluble fertilizer with a nitrogen-phosphorus-potassium ratio of 30-10-10, fertilizer pump speed 120L / h, for 15 minutes" and adds a baseline pump speed of 20L / h. If the thermal environment has the highest priority, commands such as "Turn on the fan, target speed 800rpm" or "Start the spray cooling system" are generated according to the difference between the canopy air temperature or saturated water vapor pressure difference and the target threshold, based on the corresponding proportional relationship (e.g., fan speed is proportional to the temperature difference). This enables differentiated fertilization, pesticide application, and environmental control strategies for different micro-areas.

[0075] Step S5: Control the execution of intelligent agricultural machinery equipment.

[0076] The generated instructions are converted via protocol and sent to execution terminals such as smart irrigation valves, fertilizer applicators, and environmental control cabinets through a 4G network. After the equipment performs its operations, it transmits back data such as actual irrigation duration, flow rate, fertilizer application amount, and operating status, which are then bound and stored with the original decision instructions to form a traceable decision execution record for subsequent model optimization.

[0077] Example 3 This embodiment provides a method for constructing a smart farm, which is applied to rice-growing areas.

[0078] Step S1: Multi-source environmental data acquisition.

[0079] The paddy field was divided into several micro-plots according to the location of the field and the water inlet. A water level gauge was deployed in each plot to measure the water depth and convert it into soil volumetric water content. Soil conductivity sensors and canopy temperature and humidity sensors were also deployed, and water metering equipment was installed to record the inflow and outflow of water.

[0080] Step S2: Environmental index calculation and correction.

[0081] The rice growth cycle is divided into four stages: greening stage, tillering stage, field drying stage, heading and flowering stage, and grain filling and ripening stage. The preset parameters are as follows: Tillering stage (15-35 days after transplanting): The optimum soil moisture content is set at 100% saturated moisture content, with an effective moisture content range of ±5% (corresponding to changes in water depth). The target electrical conductivity is set at 1.0 mS / cm, and the scale parameter is 0.15. The optimum saturated vapor pressure difference is set at 1.2 kPa, with an effective control range of ±0.5 kPa.

[0082] During the field drying period (36-45 days): the optimum soil moisture content is adjusted to 70% of field capacity (dry state), with an effective moisture content range of ±3%. The target electrical conductivity is set at 1.2 mS / cm, and the scale parameter is 0.2. The optimum saturated vapor pressure difference is set at 1.5 kPa, with an effective control range of ±0.6 kPa.

[0083] During the heading and flowering stage (46–65 days): the optimum soil moisture content was restored to 100%, with an effective moisture content range of ±4%. The target electrical conductivity was set at 1.3 mS / cm, and the scale parameter was 0.15. The optimum saturated vapor pressure difference was set at 1.3 kPa, with an effective control range of ±0.4 kPa.

[0084] Taking a plot of land in the tillering stage as an example: the measured soil volumetric water content is 95% (the water layer is shallow), the soil electrical conductivity is 0.9 mS / cm, the canopy air temperature is 30℃, the canopy air relative humidity is 70%, and the calculated actual saturated water vapor pressure difference is about 1.4 kPa.

[0085] Based on the preset calculation model, the following sub-indices were calculated: the water suitability index was 0, the nutrient balance index was approximately 0.80, and the environmental coordination index was 0.6.

[0086] The critical threshold for environmental coordination is set at 0.6. The current environmental coordination index is equal to the critical value, so no correction is needed. If it falls below the critical value, adjustments will be made according to the correction model, with correction intensity coefficients of 0.2 and 0.15 respectively.

[0087] The fusion model employs a weighted geometric mean function, with the weights of the tillering stage water suitability index, nutrient balance index, and environmental coordination index preset to 0.5, 0.3, and 0.2, respectively. Calculations show that the comprehensive environmental state index is 0, indicating extremely poor environmental conditions and an urgent need for water regulation.

[0088] Step S3: Output growth regulation priority.

[0089] The comprehensive environmental state index and its sub-indices are input into a neural network, and the output probability distribution is, for example, [0.85, 0.10, 0.05], clearly indicating that water regulation is the primary task.

[0090] Step S4: Generate specific farmland collaborative management decision instructions.

[0091] Based on the priority results, the target control factor is moisture. The real-time variable is soil volumetric water content, the target threshold is 100% saturated water content, and the difference is 5 percentage points. Considering the characteristics of flood irrigation, the operating time is calculated to be 52 minutes, based on a ratio of 10 minutes of operation time for every 1 percentage point difference, plus a baseline time of 2 minutes. The command generated is: "Open the inlet valve of plot A, irrigate for 52 minutes, and restore the target water depth to saturation."

[0092] For plots with high nutrient priority, based on the difference between soil electrical conductivity and the target threshold, a fertilizer application pump rate of 8 L / h is applied for every 0.1 mS / cm difference (actual proportionality coefficient 80 (L / h) / (mS·cm)). -1 The system generates commands such as "Top-dress with tillering fertilizer (urea + potassium chloride), fertilizer pump speed 80L / h, for 20 minutes" by adding a baseline pump speed. If the air and heat priority is high (e.g., high temperature during the heading stage), the system generates commands such as "Start canopy spraying for cooling, for 15 minutes" based on the difference between the canopy air temperature and the target threshold, with a ratio of 5 minutes of spraying time for every 1°C difference.

[0093] Step S5: Control the execution of intelligent agricultural machinery equipment.

[0094] Commands are sent to smart valves, fertilizer applicators, and spraying equipment via LoRa or 4G. After execution, the water level gauge confirms the water depth, the flow meter verifies the fertilizer application rate, and the data is stored in a historical database.

[0095] Example 4 This embodiment provides a method for constructing a smart farm, which is applied to sweet potato growing areas.

[0096] Step S1: Multi-source environmental data acquisition.

[0097] In the sweet potato field, micro-plots were divided according to the terrain (slope top, slope middle, and slope bottom). Soil temperature and humidity sensors, soil conductivity sensors, and canopy temperature and humidity sensors were deployed in each plot. Rain gauges and anemometers were also installed to assist in the calculation of the environmental compatibility index.

[0098] Step S2: Environmental index calculation and correction.

[0099] The sweet potato growth period is divided into the rooting and branching stage, the vegetative growth stage, the tuber enlargement stage, and the maturity stage. The preset parameters are as follows: Rooting and branching period (15-30 days after planting): Optimal soil moisture content is 70% of field capacity (about 20% for sandy loam), with an effective moisture content range of ±4%; target electrical conductivity is 1.0 mS / cm, with a scale parameter of 0.2; optimal saturated vapor pressure difference is 1.2 kPa, with an effective control range of ±0.5 kPa.

[0100] During the peak foliage growth period (31-60 days): the optimal soil moisture content is 75% of field capacity (approximately 22%), with an effective moisture content range of ±5%; the target electrical conductivity is 1.2 mS / cm, with a scale parameter of 0.2; the optimal saturated vapor pressure difference is 1.4 kPa, with an effective control range of ±0.6 kPa.

[0101] During the tuber enlargement period (61~100d): the optimal soil moisture content is 65% of field capacity (approximately 19%), with an effective moisture content range of ±3% (strictly controlled); the target electrical conductivity is 1.5 mS / cm (with appropriate increase in potassium ratio), and the scale parameter is 0.15; the optimal saturated vapor pressure difference is 1.3 kPa, with an effective control range of ±0.4 kPa.

[0102] Taking a plot of land at the bottom of a slope during the tuber enlargement period as an example: the measured soil volumetric water content is 18%, the soil electrical conductivity is 1.4 mS / cm, the canopy air temperature is 26℃, and the canopy air relative humidity is 60%. The actual saturated water vapor pressure difference is calculated to be approximately 1.6 kPa.

[0103] Based on the preset calculation model, the following sub-indices were calculated: the water suitability index was 0.67, the nutrient balance index was approximately 0.80, and the environmental coordination index was 0.25.

[0104] The critical threshold for environmental coordination is set at 0.6. The current environmental coordination index is far below the threshold and needs to be corrected according to the modified model. The correction intensity coefficient for water stress is set at 0.4, and the correction intensity coefficient for nutrient availability is set at 0.3. The calculated corrected water suitability index is 0.53, and the corrected nutrient balance index is 0.695.

[0105] The fusion model employs a weighted geometric mean function, with the weights of the moisture suitability index, nutrient balance index, and environmental coordination index during the tuber enlargement period preset to 0.5, 0.3, and 0.2, respectively. The calculated comprehensive environmental state index is approximately 0.494.

[0106] Step S3: Output growth regulation priority.

[0107] The comprehensive environmental state index and its sub-indices are input into the neural network, and the output probability distribution is, for example, [0.30, 0.25, 0.45], which shows that the control of the gas and heat environment has the highest priority.

[0108] Step S4: Generate specific farmland collaborative management decision instructions.

[0109] Based on the priority results, the target control factor is the atmospheric and thermal environment. The real-time variable is the saturated vapor pressure difference, with the target threshold being the optimum value of 1.3 kPa and the difference being 0.3 kPa. Using a ratio of 1 minute of spraying time for every 0.1 kPa difference (actual proportionality coefficient 10 min / kPa), and adding a baseline duration of 2 minutes, the spraying duration is calculated to be 5 minutes. The command generated is: "Start the spray system for 5 minutes to reduce the canopy saturated vapor pressure difference."

[0110] If other plots have a high nutrient priority, then based on the difference between soil electrical conductivity and the target threshold, a precise fertilization instruction is generated, such as "apply high-potassium water-soluble fertilizer, pump speed 100L / h, for 12 minutes," with each 0.1 mS / cm difference corresponding to a fertilization pump speed of 10L / h. If water has a high priority, then a drip irrigation instruction is generated based on the difference between soil volumetric water content and the target threshold, with each 1% difference corresponding to a drip irrigation duration of 15 minutes.

[0111] Step S5: Control the execution of intelligent agricultural machinery equipment.

[0112] Commands are sent wirelessly to the intelligent spraying system, fertilizer applicator, and drip irrigation valves. After execution, the system verifies the effect through sensor feedback, and the data is used for model optimization.

[0113] Comparative Example Based on the technical solutions of Example 1 (greenhouse tomatoes), Example 2 (maize field), Example 3 (rice), and Example 4 (sweet potato), corresponding comparative experimental groups were set up. Each comparative experimental group adopted a conventional management model relying on grower experience, specifically as follows: Irrigation management: Irrigation is carried out based on a fixed schedule or by judging by human senses (such as soil moisture level, leaf morphology), lacking accurate perception and response to the spatiotemporal variability of soil moisture.

[0114] Nutrient management: Based on experience and knowledge of crop growth stages, topdressing is carried out in a uniform manner in terms of qualitative and quantitative aspects, without making differentiated adjustments according to the real-time spatial distribution of soil nutrients.

[0115] Environmental control: By manually observing field thermometers and hygrometers, and manually turning on and off ventilation, shading, and spraying equipment, the management of each environmental factor is relatively independent, and there is a significant lag in decision-making and execution.

[0116] The following is a comparison of the effects of applying the method of this invention with conventional management on various crops. The results are shown in Tables 1-4.

[0117] Table 1 Comparison of the effects of implementing greenhouse tomatoes

[0118] Table 2 Comparison of Implementation Effects in Maize Fields

[0119] Table 3 Comparison of Implementation Effects in Rice

[0120] Table 4 Comparison of the effects of sweet potato implementation

[0121] The comparative experimental data above show that, compared with conventional management methods, the crop experimental groups applying the method described in this invention achieved significant improvements in resource utilization efficiency (water and fertilizer), crop yield and quality, production stability, and labor productivity. This verifies the effectiveness and superiority of this invention in achieving precise collaborative management through comprehensive perception of multiple environmental factors, quantitative assessment of state, and intelligent priority decision-making.

[0122] Example 5 A smart farm construction system includes: The multi-source data acquisition module is used to simultaneously collect multi-source environmental data within the farm area, including soil volumetric water content, soil electrical conductivity, canopy air temperature, and canopy air relative humidity. The environmental index calculation module is used to calculate the water suitability index, nutrient balance index, and environmental coordination index based on multi-source environmental data and through a preset calculation model. The environmental coordination index is used to correct the water suitability index and nutrient balance index through a correction model to obtain the corrected water suitability index and the corrected nutrient balance index. Finally, the corrected water suitability index, the corrected nutrient balance index, and the environmental coordination index are combined through a fusion model to generate a comprehensive environmental status index. The growth regulation decision module is used to input the comprehensive environmental state index and its sub-indices into the environmental factor coupled decision model and output the growth regulation priority; the environmental factor coupled decision model is a dedicated model based on an improved particle swarm optimization algorithm to optimize neural network parameters; The collaborative management decision-making module is used to generate specific farmland collaborative management decision-making instructions based on the control direction and intensity indicated by the growth control priority; and The intelligent execution control module is used to control intelligent agricultural machinery and equipment to execute farmland collaborative management decision-making instructions.

[0123] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0124] Therefore, this invention employs the aforementioned method and system for constructing a smart farm, building a complete solution encompassing multi-source environmental data fusion, multi-factor coupled state assessment, intelligent priority decision-making, and precise collaborative execution. This method fundamentally changes the fragmented management model of traditional single-factor threshold control, achieving precise identification and orderly regulation of environmental limiting factors.

[0125] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a smart farm, characterized in that, Includes the following steps: Step S1: Synchronously collect multi-source environmental data within the farm area; Step S2: Based on the collected multi-source environmental data, the water suitability index, nutrient balance index and environmental coordination index are calculated respectively through the preset calculation model. The water suitability index and nutrient balance index are corrected using an environmental coordination index and a correction model to obtain the corrected water suitability index and the corrected nutrient balance index. Finally, the corrected water suitability index, the corrected nutrient balance index and the environmental coordination index are combined using a fusion model to generate the comprehensive environmental state index. Step S3: Input the comprehensive environmental state index and its sub-indices into the environmental factor coupled decision model, and output the growth regulation priority; Step S4: Generate specific farmland collaborative management decision instructions based on the control direction and intensity indicated by the growth control priority; Step S5: Control the intelligent agricultural machinery and equipment to execute farmland collaborative management decision commands.

2. The method for constructing a smart farm according to claim 1, characterized in that, In step S1, the multi-source environmental data specifically includes: soil volumetric water content. Soil electrical conductivity Canopy air temperature and canopy air relative humidity .

3. The method for constructing a smart farm according to claim 2, characterized in that, In step S2, the specific calculation model is as follows: Moisture suitability index The calculation formula is: ； in, This represents the optimal soil moisture content for the current growth stage of the crop. The permissible effective water content range; Nutrient balance index The calculation formula is: ； in, The target conductivity for the current crop growth stage. This is a scale parameter characterizing tolerance. Environmental Coordination Index The calculation formula is: ； ； in, Based on the collected canopy air temperature and relative humidity The calculated actual saturated water vapor pressure difference For the optimal saturated water vapor pressure difference, To effectively regulate the scope.

4. The method for constructing a smart farm according to claim 3, characterized in that, In step S2, the model is specifically corrected as follows: Corrected Moisture Suitability Index ,in, This is the water stress correction intensity coefficient, with a value ranging from 0.1 to 0.

5. This represents the critical threshold for environmental coordination. Corrected nutrient balance index ,in, The intensity coefficient for nutrient availability is adjusted, with a value ranging from 0.05 to 0.

3. Among them, the critical threshold The value is 0.

6.

5. The method for constructing a smart farm according to claim 4, characterized in that, In step S2, the fusion model is specifically a weighted geometric mean function and an environmental comprehensive state index. The calculation formula is: ； Among them, the weighting coefficient , , These are fixed values ​​preset based on the crop's vegetative or reproductive growth stages.

6. The method for constructing a smart farm according to claim 1, characterized in that, In step S3, the environmental factor coupled decision model is a specialized model based on an improved particle swarm optimization algorithm to optimize neural network parameters. Its construction and inference process includes the following steps: a) Model structure definition: Construct a feedforward neural network as the main decision network, whose input layer nodes correspond to input features, including the comprehensive environmental state index. Corrected Moisture Suitability Index Corrected nutrient balance index Environmental Coordination Index and crop growth period coding The output layer nodes correspond to the growth regulation priority. The dimension; the number of hidden layers is 1 to 3; b) Parameter optimization: An improved particle swarm optimization algorithm is used to globally optimize the connection weights and bias parameters of the main decision network; the improvement is reflected in the fitness function design, the formula of which is: ； in, Encoding the historically optimal control decision, For the network's predicted output, To execute based on historical data statistics The estimated resource consumption required for the corresponding decision-making and To balance the weights; c) Algorithm execution: The particle swarm size is set to 50 to 200, and the particle velocity update formula incorporates an inertial weight that is dynamically adjusted based on fitness ranking. , The initial value is 0.9, and the final value at the end of the iteration is 0.4, which is the acceleration constant. ; d) Model inference: Initialize the main decision network using the optimized parameters, feed real-time input features into the network, and calculate the growth regulation priority through forward propagation. .

7. The method for constructing a smart farm according to claim 6, characterized in that, In step S3, growth regulation priority The output process is as follows: The output layer of the main decision network consists of three nodes, each outputting the initial score for water regulation. Initial score of nutrient regulation Initial score of thermal environment regulation ; Subsequently, the three initial scores are normalized into a probability distribution using the Softmax function, calculated as follows: ； in, ; Growth regulation priority That is, the probability distribution vector The category corresponding to its maximum value is determined to be the most urgent limiting factor that needs to be regulated.

8. The method for constructing a smart farm according to claim 7, characterized in that, In step S4, specific farmland collaborative management decision instructions are generated, and the following operations are performed: Step S41: Determine the priority of growth regulation The category of target regulatory factors corresponding to the maximum probability in the middle ; Step S42: Based on the target regulatory factor category Read the corresponding real-time sensor data variables from the preset control parameter mapping table. With target threshold The mapping relationship is: if If it is water, then , t If it is nutrients, then , If it is a hot environment, then , t ; Step S43: Calculate the difference And through the corresponding linear control function in the mapping table Calculate the specific equipment control parameter values. ;in This is the proportionality coefficient. As the baseline value, The specific control parameters output vary depending on the device. These represent irrigation duration, fertilizer pump speed, target fan speed, or target shade net opening, respectively.

9. A smart farm construction system, characterized in that, A method for constructing a smart farm as described in any one of claims 1-8, comprising: The multi-source data acquisition module is used to simultaneously collect multi-source environmental data within the farm area, including soil volumetric water content, soil electrical conductivity, canopy air temperature, and canopy air relative humidity. The environmental index calculation module is used to calculate the water suitability index, nutrient balance index, and environmental coordination index based on multi-source environmental data and through a preset calculation model. The environmental coordination index is used to correct the water suitability index and nutrient balance index through a correction model to obtain the corrected water suitability index and the corrected nutrient balance index. Finally, the corrected water suitability index, the corrected nutrient balance index, and the environmental coordination index are combined through a fusion model to generate a comprehensive environmental status index. The growth regulation decision module is used to input the comprehensive environmental state index and its sub-indices into the environmental factor coupled decision model and output the growth regulation priority; the environmental factor coupled decision model is a dedicated model based on an improved particle swarm optimization algorithm to optimize neural network parameters; The collaborative management decision-making module is used to generate specific farmland collaborative management decision-making instructions based on the control direction and intensity indicated by the growth control priority; and The intelligent execution control module is used to control intelligent agricultural machinery and equipment to execute farmland collaborative management decision commands.

Images