Intelligent agricultural big data system

Abstract

The invention discloses an intelligent agricultural big data system, and belongs to the field of agricultural big data, and the system comprises an environment monitoring module, a sensing layer deploys a sensor to collect original data, and a transmission layer is provided with an edge gateway which is connected with the sensor and carries out the transmission of the original data; the equipment layer is used for regulating and controlling irrigation, ventilation and light supplement schemes, and the control layer is used for optimizing MQTT message middleware to control the agricultural Internet of Things system; according to the data management module, a data collection layer obtains original data, a data processing layer processes the original data according to a three-layer data cleaning mechanism, and a data storage layer stores the processed data; according to the system management module, the data application layer constructs an interface covering Web, a mobile terminal and AR equipment, the data display layer generates a soil moisture thermodynamic diagram, a crop growth distribution diagram and an equipment operation state diagram, and intelligent management of agricultural production is ensured. The advanced Internet of Things technology is adopted, and the accuracy and the real-time performance of data acquisition are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of agricultural big data, and particularly relates to a smart agricultural big data system. BACKGROUND

[0002] Under the background of global population growth and increasing pressure on resources and the environment, how to effectively improve agricultural production efficiency and build a food security guarantee system has become a strategic issue of common concern for the international community. The deep integration of modern information technology and agricultural production has given birth to a new production model. This innovative system, known as smart agriculture, is gradually changing the traditional way of operating agriculture. By integrating sensor networks, data analysis platforms, and intelligent decision-making systems, agricultural production activities can be managed and optimized throughout the entire process, which not only improves resource utilization efficiency, but also provides technical support for the sustainable development of the agricultural ecosystem.

[0003] As the technical cornerstone of the smart agriculture system, the large-scale deployment of Internet of Things devices in the field has built a dynamic monitoring network. Distributed sensor devices in the cultivated area continuously capture key data such as soil moisture, microclimate parameters, and crop physiological indicators, and form visual decision-making suggestions through cloud algorithm processing. Agricultural producers can accurately develop water and fertilizer application plans and timely carry out pest control operations. This data-driven management model is reshaping the way modern agriculture operates.

[0004] The evolution of the domestic smart agriculture technology system presents a catch-up situation. Since the implementation of the 2015 Agricultural and Rural Information Development Plan, the policy guidance system has gradually improved, with the central government's cumulative investment of over 12 billion yuan to support the construction of digital agriculture pilot projects. Market participants show a diversified feature, with not only the Ali Cloud ET Agricultural Brain building a digital service platform for the entire supply chain, but also the Chinese Academy of Agricultural Sciences developing high-throughput collection devices for crop phenotypes. Practice shows that the soil moisture monitoring system based on narrowband Internet of Things can increase water and fertilizer utilization by 18%, but compared with the average digital penetration rate of 40% in industrialized farms in Europe and the United States, the application efficiency of domestic technology has not been fully released, and the current smart agriculture platform construction is facing multiple technical constraints. SUMMARY

[0005] In view of the above deficiencies in the prior art, the smart agricultural big data system provided by the present application solves the problem of unreasonable application of agricultural big data and low efficiency in the prior art.

[0006] In order to achieve the above-mentioned application purposes, the technical scheme adopted by the present application is as follows: a smart agricultural big data system, comprising: an environment monitoring module, which collects raw data by deploying sensors through a perception layer and transmits the raw data by setting an edge gateway to connect the sensors through a transmission layer; The device control module controls irrigation, ventilation and light supplementing scheme through the device layer, and optimizes the MQTT message middleware to control the agricultural Internet of Things system through the control layer. The data management module obtains original data through the data acquisition layer, processes the original data according to a three-layer data cleaning mechanism through the data processing layer, and stores the processed data through the data storage layer. The system management module constructs an interface covering Web, mobile terminal and AR device through the data application layer, generates a soil moisture thermal map, a crop growth distribution map and a device running state map through the data display layer, and ensures intelligent management of agricultural production.

[0007] Further, the environmental monitoring module includes a perception layer and a transmission layer. The perception layer is provided with sensors, including temperature sensors, humidity sensors and carbon dioxide concentration sensors, and the sensors are connected with edge computing nodes through RS-485 bus or LoRa wireless technology. The edge computing nodes perform preprocessing operations on the data collected by the agricultural special sensors based on the ESP32 dual-core architecture. The preprocessing operation method is as follows: running the FreeRTOS real-time system to process sensor polling collection, executing floating-point operation based on the CMSIS-DSP library, implementing Kalman filter noise reduction on the original data collected by the sensors, detecting abnormal gradient changes by using the sliding window algorithm, filtering environmental noise in real time by Kalman filter, and completing data preprocessing operation. The transmission layer includes an edge gateway and a protocol conversion submodule. The edge gateway is connected with the edge nodes. The edge gateway adopts the MQTT protocol and is loaded with a high-performance industrial-grade processor based on the ARM Cortex-A53 architecture. The security transmission mechanism of the protocol conversion submodule adopts the SM4-CTR encryption mode, generates a message authentication code in combination with the HMAC-SM3, and protects the root key through a hardware-level security chip based on the PUF physical unclonable function.

[0008] Further, the device control module includes a device layer and a control layer. The device layer includes an irrigation submodule, a ventilation submodule and a light supplementing submodule. The irrigation submodule controls the intelligent water pump by using the adaptive PID control algorithm, dynamically adjusts the water pressure according to the feedback data of the soil humidity sensor, and performs hierarchical irrigation on crops. The ventilation submodule is provided with a variable frequency speed regulation fan, which is linked with the wind speed sensor and the carbon dioxide detector to maintain the ventilation state of the greenhouse. The light supplementing submodule is provided with a PWM light adjusting controller, which generates a light supplementing scheme containing 660nm red light and 450nm blue light according to the light period requirement of crops. The control layer sets an MQTT message middleware, key instructions are transmitted in QoS 2 level, an 8-byte binary packet is designed, a security system adopts X.509 certificate bidirectional authentication, a device end stores an SM2 private key through a Secure Element chip, and a control instruction adopts an ASN.1 encoding format signature.

[0009] Further, the data management module includes a data acquisition layer, a data processing layer and a data storage layer. The data acquisition layer is equipped with a dual-core ARM processor and an FPGA acceleration chip, accesses traditional equipment through Modbus RTU, connects distributed sensor nodes by using LoRaWAN, and supports 5G NR transmission of unmanned aerial vehicle inspection data. The data processing layer adopts a three-layer data cleaning mechanism driven by a dynamic rule engine: the first layer filters abnormal values of original data, the second layer adopts a sliding window interpolation method to repair time sequence disorder, adopts a Laplace criterion to eliminate gross errors, converts data into a unified dimension, aligns time series data to a standard timestamp, converts spatial data to a WGS84 coordinate system, and the third layer identifies device failure modes and labels based on a random forest algorithm; The data storage layer adopts a hybrid architecture: Redis cluster caches the latest sensor data, each data point saves original values, calibrated values and quality marks, and storage contents include: multispectral images shot by an unmanned aerial vehicle, a pest picture library identified by AI, and a farm machine operation track log.

[0010] Further, the data acquisition layer sets an intelligent cache mechanism, specifically: In response to network delay exceeding 500 milliseconds, switching to a local storage mode, and adopting an optimized ring buffer structure to save data.

[0011] Further, the system management module includes a data application layer and a data display layer. The data application layer includes a disaster warning submodule, an AR auxiliary submodule and a decision support submodule; the disaster warning submodule constructs a Bayesian network inference engine, converts multidimensional parameters of meteorological data, device states and crop growth stages into a conditional probability matrix, and performs disaster warning; the AR auxiliary submodule integrates a SLAM real-time positioning and map construction algorithm, performs device identification by ORB feature point matching, and constructs a cross-platform augmented reality interface by using a WebXR standard; the decision support submodule converts water and fertilizer costs, environmental constraints and yield prediction parameters into a Pareto frontier solution set by using a multi-objective optimization algorithm, and outputs a multi-dimensional decision scheme by using an NSGA-II genetic algorithm; The data display layer includes an intelligent analysis submodule, a visualization submodule and an intelligent report submodule. The intelligent analysis submodule is based on an environment prediction model constructed by an LSTM neural network, and includes prediction models for crop yield, soil moisture and agricultural yield. The visualization submodule uses Vue3+WebGL technology to construct a three-dimensional digital farm, and displays soil moisture thermal maps, crop growth distribution maps and device operation state maps in layers. The intelligent report submodule integrates an agricultural semantic analysis engine, uses a bidirectional LSTM model to extract key operation events from device logs, and generates standardized reports containing summaries of agricultural operations and correlation analysis of environmental factors.

[0012] Further, the workflow of the environment prediction model is as follows: A1, obtain the original data collected by the sensor, filter the outliers from the original data based on the Laplace criterion, delete the outliers and repair them by linear interpolation, and clean the repaired data by random forest to obtain the cleaned data; A2, decompose the cleaned data by STL to obtain the trend item and the seasonal item, input the trend item into the ARIMA model to obtain the trend item prediction result; A3, perform feature engineering on the cleaned data, set a sliding window statistic, calculate the mean and variance, obtain meteorological factors from temperature and humidity according to an empirical formula, and establish a feature vector according to the meteorological factors, the mean and the variance; A4, input the historical data and the feature vector into the LSTM neural network to obtain the spatio-temporal feature; A5, weight and fuse the trend item prediction result, the seasonal item and the spatio-temporal feature to obtain the output result of the environment prediction model.

[0013] Further, in A5, the expression of the output result of the environment prediction model is as follows: In the formula, is a full connection layer, is a spatio-temporal feature, is a trend item prediction result, is a seasonal item, , and are dynamic weights. In the formula, is the mean absolute error of the LSTM neural network on the validation set, is the mean absolute error of the ARIMA model on the validation set, is the mean absolute error of the STL decomposition method on the validation set.

[0014] Further, the visualization submodule adopts the WebGL technology to construct a full-scene agricultural operation interface, and the interaction mode is as follows: the instantiation rendering technology is adopted to optimize the farmland model loading efficiency, and the soil moisture thermal map and device distribution topology are dynamically generated in combination with the quadtree spatial index.

[0015] The present application has the following advantages: (1) The present application provides a smart agricultural big data system, the environment monitoring module adopts advanced Internet of Things technology to clean, convert and standardize the data, and at the same time provides data quality monitoring and data version management functions, improves the accuracy and real-time performance of data acquisition, and builds a high-availability and high-extensibility distributed system through modular components, realizes real-time monitoring and accurate regulation of agricultural production environment and crop growth conditions.

[0016] (2) The device control module and the data management module of the present application use big data analysis and artificial intelligence algorithms to improve the data processing and decision support capabilities, the device control module covers the access of water and fertilizer integrated machines, irrigation systems, ventilation equipment and other agricultural equipment, and real-time control of the device running state, thereby providing scientific decision support for agricultural production, optimizing resource allocation schemes and improving production efficiency. The data management module uses MySQL to store business data, Redis to cache hot data, and TDengine to store time series data, and at the same time has a data backup and recovery mechanism to ensure the safety and reliability of the data.

[0017] (3) The present application builds a multi-layer superimposed visualization sand table based on the WebGL technology through the system management module: the underground 20cm soil moisture thermal map, the crop canopy NDVI index layer and the historical pest and disease distribution layer are fused on the farmland two-dimensional base map, and the sliding time axis is supported to observe the evolution trend of the agricultural conditions, so as to reduce the learning cost of users and promote the popularization of the system. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 It is a schematic diagram of a smart agricultural big data system of the present application. DETAILED DESCRIPTION

[0019] The specific embodiments of the present application are described below to facilitate those skilled in the art to understand the present application, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions utilizing the concept of the present application are within the scope of protection.

[0020] As shown in Figure 1 in an embodiment of the present application, a smart agricultural big data system comprises: The environment monitoring module collects original data through sensors deployed by the perception layer and transmits the original data through the transmission layer by setting an edge gateway to connect the sensors; The device control module controls irrigation, ventilation and light supplementing schemes through the device layer and optimizes the MQTT message middleware to control the agricultural Internet of Things system through the control layer; The data management module obtains original data through the data acquisition layer, processes the original data according to a three-layer data cleaning mechanism through the data processing layer, and stores the processed data through the data storage layer. The system management module constructs an interface covering Web, mobile terminals and AR devices through the data application layer, generates a soil moisture condition thermal map, a crop growth distribution map and a device operation state map through the data display layer, and ensures intelligent management of agricultural production.

[0021] The environment monitoring module includes a perception layer and a transmission layer; The perception layer is provided with sensors, including temperature sensors, humidity sensors and carbon dioxide concentration sensors; in this embodiment, the perception layer collects environmental parameters such as temperature and humidity, light, CO2 concentration and soil moisture condition in real time, the temperature sensor adopts an industrial-grade packaged DS18B20 chip, the stainless steel shell of which can resist fertilizer corrosion and cooperate with an IP68 waterproof design to be able to be buried in soil or hung in a humid breeding house for a long time; the humidity sensor selects a DHT22 double-probe version, which can not only monitor air humidity fluctuation but also obtain soil moisture condition data in real time through an extension probe; the optical fiber sensor BH1750 adopts a digital output mode to avoid the attenuation problem of traditional analog signals in long-distance transmission, and the carbon dioxide concentration sensor MH-Z19 realizes high-precision detection in the range of 400-5000ppm through a non-dispersive infrared (NDIR) technology, and its automatic calibration function can avoid measurement drift caused by long-term use. The perception layer also supports access of various Internet of Things devices such as weather stations, soil sensors and water quality monitors, and is compatible with GB28181 protocol cameras to realize video monitoring and image acquisition.

[0022] The sensors are connected with edge computing nodes through RS-485 buses or LoRa wireless technologies, the edge computing nodes perform preprocessing operations on the data collected by the agricultural special-purpose sensors based on an ESP32 dual-core architecture, and the method of the preprocessing operations is specifically as follows: a FreeRTOS real-time system is run to process sensor polling collection, a floating-point operation based on a CMSIS-DSP library is executed, Kalman filter denoising is implemented on the original data collected by the sensors, a sliding window algorithm is used to detect abnormal gradient changes, environmental noise is filtered in real time through Kalman filtering, and the data preprocessing operations are completed. In this embodiment, the edge computing node adopts a microcontroller with dual-core processing capability such as ESP32 to complete basic data collection and perform preliminary data preprocessing. For example, the temperature data is subjected to moving average filtering to eliminate transient interference, the original ADC reading is converted into a standard measurement unit, and abnormal values exceeding a reasonable range are marked or removed.

[0023] The transmission layer includes an edge gateway and a protocol conversion submodule. The edge gateway is connected with the edge node. The edge gateway adopts the MQTT protocol and is equipped with a high-performance industrial processor based on the ARM Cortex-A53 architecture. In this embodiment, the edge gateway serves as the hub device of the transmission layer and is equipped with a dual-core ARM processor and an FPGA acceleration chip. In the precise irrigation scenario, the gateway is built-in with a fuzzy control algorithm, which can dynamically adjust the valve opening in combination with the local soil permeability data to effectively avoid irrigation waste caused by soil compaction.

[0024] The security transmission mechanism of the protocol conversion submodule adopts the SM4-CTR encryption mode and generates a message authentication code in combination with HMAC-SM3. The key management is protected by a root key through a hardware-level security chip based on a PUF physical unclonable function.

[0025] In this embodiment, the protocol conversion submodule realizes Modbus RTU / TCP protocol analysis based on the libmodbus open source library and constructs a JSON Schema in line with the characteristics of agricultural data in combination with the Jansson library to ensure the semantic consistency of industrial device data to the cloud platform.

[0026] The device control module includes a device layer and a control layer. The device layer includes an irrigation submodule, a ventilation submodule, and a light supplement submodule. The irrigation submodule controls an intelligent water pump using an adaptive PID control algorithm, dynamically adjusts water pressure according to the feedback data of a soil humidity sensor, performs hierarchical irrigation on crops, and realizes hierarchical precise irrigation of a slope orchard. The ventilation submodule sets a variable frequency speed regulation fan, which is linked with a wind speed sensor and a carbon dioxide detector to maintain the ventilation state of the greenhouse. The light supplement submodule sets a PWM light control controller to generate a light supplement scheme containing a 660nm red light and a 450nm blue light according to the light period requirement of crops. In this embodiment, the sub-modules of the device layer access the gateway through RS-485 bus or ZigBee wireless network, forming a device cluster with edge computing capability. At the device control layer, the intelligent water pump dynamically adjusts the PID parameters through the Ziegler-Nichols tuning method, and combines the integral anti-windup mechanism to avoid irrigation overshoot; the frequency conversion fan adopts FOC technology, realizes three-phase current decoupling through Clarke-Park transformation, and cooperates with the PI controller to accurately adjust the speed; the light supplementing sub-module uses PWM and constant current drive composite dimming technology, realizes ±5nm precision spectrum control based on I2C bus instructions, and provides reliable automatic execution foundation for agricultural production.

[0027] The control layer sets up MQTT message middleware, and key instructions are transmitted in QoS 2 level. The designed 8-byte binary packet (including device address, operation code and parameter value) improves transmission efficiency by 60% compared with JSON format; the security system adopts X.509 certificate two-way authentication, the device end stores SM2 private key through Secure Element chip, and the control instruction adopts ASN.1 encoding format signature, which fully prevents man-in-the-middle attack and ensures safe and reliable operation of agricultural Internet of Things system.

[0028] In this embodiment, the control layer builds the digital nerve center of the intelligent agricultural system, and realizes precise control through hierarchical communication architecture. The bottom layer device uses MQTT protocol with multi-level topic structure (such as "farm / some base / xx number greenhouse / irrigation / status") to upload state data efficiently, and the control gateway deploys RabbitMQ image queue cluster to ensure service high availability. The protocol level specially optimizes the binary encoding format of agricultural control instructions, encapsulates irrigation time, light intensity and other parameters into compact data packets, and significantly reduces transmission flow.

[0029] The data management module includes a data acquisition layer, a data processing layer and a data storage layer; Among them, the data acquisition layer is equipped with dual-core ARM processor and FPGA acceleration chip, accesses traditional equipment through Modbus RTU, connects distributed sensor nodes through LoRaWAN, and supports 5G NR transmission of unmanned aerial vehicle inspection data; In this embodiment, the data acquisition layer uses an industrial-grade edge computing gateway as the field center, equipped with dual-core ARM processor and FPGA acceleration chip, and has multi-protocol compatibility. To solve the problem of unstable farmland network, an intelligent cache mechanism is designed, which is specifically: In response to network delay exceeding 500 milliseconds, switch to local storage mode, adopt optimized ring buffer structure to save data, and ensure 72 hours of data without loss. In the smart pasture scene, the collector integrates an RFID module, associates livestock ear tag information with environmental data in real time, and establishes individual growth archives. For security protection, the video stream is encrypted end-to-end using the national SM9 algorithm, providing a reliable data foundation for subsequent data processing.

[0030] The data processing layer adopts a three-layer data cleaning mechanism driven by a dynamic rule engine: the first layer filters abnormal values of raw data, the second layer uses a sliding window interpolation method to repair timing disorder, uses the Laplace criterion to remove gross errors, converts data to a unified dimension, aligns time series data to a standard timestamp, and converts spatial data to the WGS84 coordinate system, and the third layer identifies device failure modes based on a random forest algorithm and labels them. In this embodiment, the data processing layer is the core processing link of agricultural big data, and builds an efficient conversion system from raw data to business insights. A three-layer data cleaning mechanism driven by a dynamic rule engine is used to complete the unified processing and analysis of multi-source data.

[0031] The data storage layer adopts a hybrid architecture: Redis cluster caches the latest sensor data, each data point saves the original value, calibrated value and quality mark, and the storage content includes: multispectral images taken by drones, pest and disease picture library identified by AI, and farm machinery operation trajectory log.

[0032] In this embodiment, the data storage layer builds a multi-dimensional and three-dimensional warehouse of agricultural data, and meets the data access needs of different scenarios through a hybrid storage architecture. Redis cluster is deployed in sentinel mode, and caches the last 72 hours of high-frequency access data: including real-time environmental monitoring values, device warning states, user session information, etc. MySQL database uses a sharding strategy, hashes by farm ID, deploys MGR cluster for each shard to ensure high availability, and stores structured data such as device metadata, user permissions, and farm records. The time series database uses InfluxDB.

[0033] The system management module includes a data application layer and a data display layer; The data application layer includes a disaster warning submodule, an AR assistance submodule, and a decision support submodule; The disaster warning submodule builds a Bayesian network inference engine, converts multidimensional parameters such as weather data, device status, and crop growth stage into conditional probability matrices, and performs disaster warning; In this embodiment, the disaster warning sub-module realizes accurate and efficient disaster warning. The first-level warning (such as extreme weather warning) triggers sound and light alarm and short message notification to all relevant personnel; the second-level warning (such as device offline) generates a to-do task and pushes it to the mobile phone APP of the responsible person; and the third-level warning (such as trend parameter deviation) generates an analysis report in the system, providing scientific decision support for agricultural producers.

[0034] The AR auxiliary sub-module integrates the SLAM instant positioning and map construction algorithm, identifies the device marker through ORB feature point matching, and adopts the WebXR standard to construct a cross-platform augmented reality interface. In this embodiment, the AR auxiliary sub-module realizes on-site operation guidance combined with AR technology: the technical personnel only need to wear smart glasses and swipe the device two-dimensional code, and detailed maintenance guidance video and historical maintenance records can be presented in front of them.

[0035] The decision support sub-module adopts a multi-objective optimization algorithm to convert water and fertilizer cost, environmental constraints and yield prediction parameters into a Pareto frontier solution set, and outputs a multi-dimensional decision scheme through an NSGA-II genetic algorithm. In this embodiment, the decision support sub-module innovatively applies a multi-objective optimization algorithm to convert water and fertilizer cost, environmental constraints, yield prediction and other parameters into a Pareto frontier solution set, and provides a multi-dimensional decision scheme for farmers through an NSGA-II genetic algorithm.

[0036] The data display layer includes an intelligent analysis sub-module, a visualization sub-module and an intelligent report sub-module. The intelligent analysis sub-module is based on an environmental prediction model constructed by an LSTM neural network, including a crop yield, soil moisture and agricultural yield prediction model. In this embodiment, the intelligent analysis sub-module constructs an LSTM-GRU hybrid neural network model, designs a multi-modal feature fusion mechanism in the input layer, spatially aligns the environmental time series data and the multi-spectral image feature vector, and dynamically adjusts the feature weight through an attention mechanism. The model training adopts a transfer learning strategy, pre-trains based on the PlantVillage public data set, and then fine-tunes using local collected data, so as to improve the accuracy of disease and pest identification.

[0037] In this embodiment, for the prediction of crop yield, soil moisture and agricultural yield, the application designs a dynamic prediction model that changes with data. The working process of the environmental prediction model is as follows: A1, acquire the original data collected by the sensor, filter the abnormal values from the original data based on the Laplace criterion, delete the abnormal values and repair them by linear interpolation, and clean the repaired data by a random forest to obtain cleaned data; A2. Perform STL decomposition on the cleaned data to obtain trend terms and seasonal terms. Input the trend terms into the ARIMA model to obtain the trend term prediction results. A3. Perform feature engineering on the cleaned data, set sliding window statistics, calculate the mean and variance, use empirical formulas to derive meteorological factors from temperature and humidity, and create feature vectors based on the meteorological factors, mean, and variance. A4. Input historical data and feature vectors into the LSTM neural network to obtain spatiotemporal features. A5. Perform weighted fusion on the trend item prediction results, seasonal items, and spatiotemporal features to obtain the output results of the environmental prediction model.

[0038] In this embodiment, for the prediction model of crop yield, the raw data collected by the sensor is the crop yield.

[0039] In A1, the present invention uses the Laida criterion + sliding window + random forest to preprocess the data to obtain cleaned data. Based on the Laida criterion, abnormal data is deleted and interpolated. That is, if the following formula is satisfied, it is identified as an outlier, all outliers are deleted, and the linear interpolation method is used to repair it; Where, is the sliding window mean, For the i data samples, , is the original data, is the standard deviation; The patched data is cleaned by random forest, and the cleaned data The specific expression is: Where, K is the total number of decision trees, For the k Tree Pair Fault judgment, is the fault threshold, which is usually set to 0.3 in this embodiment; In A2, the STL decomposition is as follows: , calculate the trend terms respectively , Seasonal items and the residual ; Where, is a local weighted regression smoothing, k is a sliding window size, and the embodiment sets it to 30 days, P is a period, and the embodiment sets it to 7 days, is the observation value at the t - i moment in the time series, is the trend component at the t - i moment in the time series.

[0040] The expression of the ARIMA model is specifically as follows: In the formula, and are autoregressive coefficients, and are autoregressive coefficients, and are lag operators and p order lag operators respectively, is a d-order difference operator, is an ARIMA white noise, .

[0041] In A3, the expression of the mean and the variance is specifically as follows: In the formula, is a sliding window statistic, is the t moment in the time series, i is the moment in the sliding window, is the sliding window mean.

[0042] The expression of the meteorological factor is specifically as follows: In the formula, is temperature, is humidity; The characteristic vector is established.

[0043] In A5, the expression of the output result of the environmental prediction model is specifically as follows: wherein, is a fully connected layer, is a spatio-temporal feature, is a trend item prediction result, is a seasonal item, , and are dynamic weights; wherein, is the mean absolute error of the LSTM neural network on the validation set, is the mean absolute error of the ARIMA model on the validation set, is the mean absolute error of the STL decomposition method on the validation set.

[0044] The visualization submodule adopts Vue3+WebGL technology to build a three-dimensional digital farm, and displays the soil moisture-thermal map, crop growth distribution map and device running state map in layers. The visualization submodule adopts WebGL technology to build a full-scene agricultural operation interface, and its interaction mode is: adopting instance rendering technology to optimize the loading efficiency of the farmland model, and combining a quadtree spatial index to dynamically generate a soil moisture-thermal map and a device distribution topology.

[0045] In this embodiment, the visualization submodule is based on Vue3+WebGL technology, adopts instance rendering technology to optimize the loading efficiency of the farmland model, combines a quadtree spatial index to dynamically generate a soil moisture-thermal map and a device distribution topology, and displays the soil moisture-thermal map, crop growth distribution map and device running state map in layers. The user can freely switch the time dimension, compare the environmental parameter curves of different growth periods, and the system automatically labels the key farming operation time points (such as fertilization and irrigation) to help analyze the actual effect of management measures.

[0046] The intelligent report submodule integrates an agricultural semantic analysis engine, adopts a bidirectional LSTM model to extract key operation events from device logs, and generates a standardized report containing farming operation summaries and environmental factor correlation analysis.

[0047] In this embodiment, the intelligent report submodule automatically generates a multi-dimensional analysis report: compares the stress resistance data of different varieties, analyzes the water and fertilizer input-output ratio, and evaluates the potential impact of climate change on yield.

[0048] In the description of the application, it needs to be understood that the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. In addition, the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implying the number of technical features indicated. Therefore, the features defined by "first", "second", "third" can explicitly or implicitly include one or more of the features.

Claims

1. A smart agricultural big data system, characterized by: include: The environmental monitoring module deploys sensors at the perception layer to collect raw data, and sets up edge gateways at the transport layer to connect to the sensors and transmit the raw data. The device control module regulates irrigation, ventilation, and lighting solutions at the device layer and controls the agricultural IoT system by optimizing the MQTT messaging middleware at the control layer. The data management module obtains raw data through the data acquisition layer, processes the raw data according to the three-layer data cleaning mechanism through the data processing layer, and stores the processed data through the data storage layer; The system management module builds an interface covering the Web, mobile terminals, and AR devices through the data application layer, and generates soil moisture heat maps, crop growth distribution maps, and equipment operation status maps through the data display layer to ensure intelligent management of agricultural production.

2. The smart agricultural big data system according to claim 1, characterized in that: The environmental monitoring module includes a perception layer and a transmission layer; The perception layer is equipped with sensors, including temperature sensors, humidity sensors, and carbon dioxide concentration sensors. The sensors are connected to the edge computing node via the RS-485 bus or LoRa wireless technology. The edge computing node preprocesses the data collected by the agricultural sensors based on the ESP32 dual-core architecture. The preprocessing method is as follows: running the FreeRTOS real-time system to process sensor polling collection, performing floating-point operations based on the CMSIS-DSP library, applying Kalman filtering to the raw data collected by the sensors for denoising, using the sliding window algorithm to detect abnormal gradient changes, and filtering environmental noise in real time through the Kalman filter to complete the data preprocessing operation. The transport layer includes an edge gateway and a protocol conversion submodule. The edge gateway is connected to the edge node and uses the MQTT protocol. It is equipped with a high-performance industrial-grade processor based on the ARM Cortex-A53 architecture. The secure transmission mechanism of the protocol conversion submodule adopts the SM4-CTR encryption mode and combines HMAC-SM3 to generate a message authentication code. Key management uses a hardware-level security chip based on the PUF physical unclonable function to protect the root key.

3. The smart agricultural big data system according to claim 1, characterized in that: The device control module includes the device layer and the control layer; The equipment layer includes an irrigation submodule, a ventilation submodule, and a light-supplementing submodule. The irrigation submodule uses an adaptive PID control algorithm to control an intelligent water pump, dynamically adjusting water pressure based on feedback from a soil moisture sensor to irrigate crops in stages. The ventilation submodule uses a variable-frequency speed-regulating fan, which works in conjunction with a wind speed sensor and a carbon dioxide detector to maintain ventilation in the greenhouse. The light-supplementing submodule uses a PWM dimming controller to generate a light-supplementing solution consisting of a ratio of 660nm red light to 450nm blue light based on the crop's photoperiod requirements. The control layer sets up MQTT message middleware, key instructions use QoS level 2 transmission, and the designed 8-byte binary data packet. The security system uses X.509 certificate two-way authentication. The device side stores the SM2 private key through the Secure Element chip, and the control instructions are signed in ASN.1 encoding format.

4. The smart agricultural big data system according to claim 3, characterized in that: The data management module includes data acquisition layer, data processing layer and data storage layer; The data acquisition layer is equipped with a dual-core ARM processor and FPGA acceleration chip, connects to traditional devices via Modbus RTU, uses LoRaWAN to connect distributed sensor nodes, and supports 5G NR to transmit drone inspection data. The data processing layer uses a three-layer data cleaning mechanism driven by a dynamic rule engine: the first layer filters outliers in the raw data; the second layer uses sliding window interpolation to repair time series errors; the Laida criterion is used to eliminate gross errors; the data is converted to a unified dimension; time series data is aligned to standard timestamps; spatial data is converted to the WGS84 coordinate system; and the third layer uses a random forest algorithm to identify and label equipment failure modes. The data storage layer adopts a hybrid architecture: the Redis cluster caches the latest sensor data, and each data point saves the original value, calibration value and quality mark. The stored content includes: multispectral images taken by drones, AI-identified pest and disease image libraries and agricultural machinery operation trajectory logs.

5. The smart agricultural big data system according to claim 4, characterized in that: The data collection layer sets up an intelligent caching mechanism, specifically: In response to a network delay exceeding 500 milliseconds, the system switches to local storage mode and uses an optimized ring buffer structure to save data.

6. The smart agricultural big data system according to claim 1, characterized in that: The system management module includes the data application layer and the data display layer; The data application layer includes a disaster warning submodule, an AR assistance submodule, and a decision support submodule. The disaster warning submodule builds a Bayesian network inference engine to convert multi-dimensional parameters such as meteorological data, equipment status, and crop growth stage into a conditional probability matrix for disaster warning. The AR assistance submodule integrates SLAM real-time positioning and map construction algorithms, identifies equipment identifiers through ORB feature point matching, and uses the WebXR standard to build a cross-platform augmented reality interface. The decision support submodule uses a multi-objective optimization algorithm to convert parameters such as water and fertilizer costs, environmental constraints, and yield prediction into a Pareto frontier solution set, and outputs a multi-dimensional decision solution using the NSGA-II genetic algorithm. The data display layer includes an intelligent analysis sub-module, a visualization sub-module and an intelligent reporting sub-module. The intelligent analysis sub-module is based on an environmental prediction model built on the LSTM neural network, including prediction models for crop yield, soil moisture and agricultural yield. The visualization sub-module uses Vue3+WebGL technology to build a three-dimensional digital farm, displaying soil moisture heat maps, crop growth distribution maps and equipment operation status maps in layers; the intelligent reporting sub-module integrates an agricultural semantic parsing engine, uses a bidirectional LSTM model to extract key operation events from equipment logs, and generates standardized reports containing agricultural operation summaries and environmental factor correlation analysis.

7. The smart agricultural big data system according to claim 6, characterized in that: The specific workflow of the environmental prediction model is as follows: A1. Obtain raw data collected by sensors, filter outliers from the raw data based on the Laida criterion, delete outliers, and use linear interpolation to patch the patched data. Clean the patched data using a random forest cleanser to obtain cleaned data. A2. Perform STL decomposition on the cleaned data to obtain trend terms and seasonal terms. Input the trend terms into the ARIMA model to obtain the trend term prediction results. A3. Perform feature engineering on the cleaned data, set sliding window statistics, calculate the mean and variance, use empirical formulas to derive meteorological factors from temperature and humidity, and create feature vectors based on the meteorological factors, mean, and variance. A4. Input historical data and feature vectors into the LSTM neural network to obtain spatiotemporal features. A5. Perform weighted fusion on the trend item prediction results, seasonal items, and spatiotemporal features to obtain the output results of the environmental prediction model.

8. The smart agricultural big data system according to claim 7, characterized in that: In A5, the output of the environmental prediction model is obtained The specific expression is: Where, is the fully connected layer, is the spatiotemporal feature, is the trend item prediction result, is the seasonal term, 、 and All are dynamic weights; Where, is the mean absolute error of the LSTM neural network on the validation set, is the mean absolute error of the ARIMA model on the validation set, is the mean absolute error of the STL decomposition method on the validation set.

9. The smart agricultural big data system according to claim 8, characterized in that: The visualization submodule uses WebGL technology to build a full-scene agricultural operation interface. Its interactive method is: using instanced rendering technology to optimize the efficiency of farmland model loading, and combining quadtree spatial indexing to dynamically generate soil moisture heat maps and equipment distribution topology.

Images