Enterprise ERP comprehensive management system and method based on big data

Abstract

The invention relates to the technical field of intelligent agriculture, and discloses an enterprise ERP comprehensive management system and method based on big data. The enterprise ERP comprehensive management method based on big data comprises the following steps: acquiring data from satellite remote sensing, a soil sensor, unmanned aerial vehicle multispectral imaging, meteorological information, agricultural material supply chain information and agricultural product sales information; carrying out cleaning, integration and feature engineering processing on the collected data; inputting the processed data into a model to obtain the yield prediction of the crops; optimizing a supply chain of the agricultural products; and constructing an agricultural product quality tracing system. According to the invention, data of each link of agricultural production can be effectively integrated, agricultural production decision making is assisted through accurate yield prediction, planting resources are reasonably planned, and the yield and quality of crops are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of smart agriculture technology, more specifically, it relates to an enterprise ERP comprehensive management system and method based on big data. BACKGROUND

[0002] In today's digital age, enterprise resource planning (ERP) systems are crucial for the efficient operation of enterprises. Especially in the agricultural field, with the development of big data technology, enterprise ERP comprehensive management systems based on big data have become the key to improving the competitiveness of agricultural enterprises.

[0003] However, the current existing agricultural enterprise ERP system has many shortcomings. For example, in the configuration of production factors, the traditional system is difficult to integrate multi-source heterogeneous data such as satellite remote sensing, soil sensors, unmanned aerial vehicle multispectral imaging, etc. These data sources are extensive and have various formats, and the traditional system cannot effectively integrate them, resulting in inaccurate analysis of the impact of land, crops, weather, etc. on production, and thus inefficient production factor configuration. At the same time, the existing system also has shortcomings in industry chain collaboration. The information flow between each link is not smooth, and cross-farm data collaborative analysis cannot be achieved, resulting in insufficient industry chain collaboration and inability to respond to market changes in a timely manner. In addition, market response lag is also a common problem. Due to the lack of effective data mining and prediction technology, enterprises are difficult to accurately grasp market dynamics and make reasonable production and sales decisions. SUMMARY

[0004] The present application provides an enterprise ERP comprehensive management system and method based on big data. The system integrates satellite remote sensing, soil sensors, unmanned aerial vehicle multispectral imaging, etc. 7 types of heterogeneous data sources to build an agricultural big data lake through the data acquisition layer; adopts an improved federated learning framework (FL-Edge) to realize cross-farm data collaborative analysis under the premise of ensuring data privacy, and improves the efficiency of model training; the data processing layer constructs an agricultural knowledge base containing crop growth models, supply chain relationship networks, and market prediction factors with the help of dynamic knowledge graph technology, supports semantic-level data correlation query; develops a yield prediction model based on TransformerXL, uses meteorological time series attention mechanism, etc. to improve prediction accuracy; uses a supply chain optimization algorithm combining mixed integer programming and deep Q network (DQN) to reduce logistics costs; constructs a blockchain-enabled agricultural product quality traceability system, uses a lightweight consensus algorithm (PoET) to improve transaction processing and traceability query efficiency. For the reliability problem of sensor networks in extreme environments, develop low-temperature self-heating sensor packaging technology and data compensation algorithm, and realize cross-base knowledge transfer through federated learning, introduce online learning mechanism and concept drift detection module, enhance the adaptability and stability of the system in extreme environments, solve the problems of inefficient production factor configuration, insufficient industry chain collaboration, and market response lag faced by agricultural enterprises.

[0005] The application provides an enterprise ERP comprehensive management method based on big data, comprising: Through satellite remote sensing, land cover and crop distribution information are obtained; soil parameter information is obtained through a soil sensor; red, green, blue, near-infrared and short-wave infrared band information of unmanned aerial vehicle multi-spectral imaging is obtained; agricultural supply chain data, agricultural product sales data, historical comprehensive data and weather information are obtained; The collected data are subjected to cleaning, integration and feature engineering processing, abnormal values are eliminated, missing values are filled, data formats are unified, and key features are extracted; Based on the preprocessed data, a TransformerXL deep learning model is used to predict the yield of different crops; A mixed integer programming, deep reinforcement learning and multi-agent reinforcement learning optimization algorithm are comprehensively used to intelligently optimize enterprise inventory management, logistics distribution and production and sales collaborative supply chain links; A blockchain technology is used to build a quality traceability system for agricultural products covering the whole production, processing and circulation chain.

[0006] In a preferred embodiment, in the data collection step, the data collection frequency of each data source is as follows: satellite remote sensing data: high-frequency collection; soil sensor data: high-frequency collection; unmanned aerial vehicle multi-spectral imaging data: regular collection; weather information: high-frequency collection; agricultural supply chain information: real-time collection; agricultural product sales information: real-time collection.

[0007] In a preferred embodiment, the data preprocessing step comprises: Data cleaning: abnormal value detection and missing value filling methods are used to improve data quality; Data integration: the data of heterogeneous data sources are converted into a unified format and are fused; Feature engineering: key features are extracted from the original data and feature selection is performed.

[0008] In a preferred embodiment, the yield prediction step uses a TransformerXL model to accurately predict the yield of crops based on massive historical data and weather and soil influencing factors.

[0009] In a preferred embodiment, the TransformerXL model comprises: Input layer: receiving time series weather information yield influencing factor data; Weather time series attention layer: focusing on the influence of different time steps of weather factors on yield through an attention mechanism; Hidden layer: further extracting and transforming features and mining complex relationships between weather and yield; Output layer: output the predicted crop yield results.

[0010] In a preferred embodiment, the supply chain optimization step employs one or more of the following optimization algorithms: Mixed Integer Programming: for inventory management optimization, improving inventory turnover rate; Deep Reinforcement Learning: for logistics distribution optimization, improving distribution efficiency; Multi-Agent Reinforcement Learning: for production and marketing coordination optimization, improving demand prediction accuracy.

[0011] In a preferred embodiment, mixed integer programming is combined with deep Q-network to achieve supply chain optimization through the following steps: Determine the state and action space; Initialize the Q-value function; Perform actions and obtain rewards; Update the Q-value function using the Q-value update formula.

[0012] In a preferred embodiment, the agricultural product quality traceability step utilizes the following key technologies: Blockchain and Internet of Things: for the production link, realize the on-chain operation of agricultural operation data, ensure the authenticity and credibility of data; Blockchain and smart contract: for the processing link, realize quality qualification certification, improve product qualification rate; Blockchain and two-dimensional code: for the circulation link, realize product information traceability query, shorten response time.

[0013] In a preferred embodiment, the blockchain adopts a lightweight consensus algorithm PoET to achieve efficient consensus through the following steps: Each node participates in consensus; Randomly select validators; The validators process transactions and reach consensus with other nodes.

[0014] In a preferred embodiment, an enterprise ERP comprehensive management system based on big data includes the following modules: Data acquisition module, used for multi-source data acquisition from satellite remote sensing, soil sensor, unmanned aerial vehicle multispectral imaging, weather, agricultural supply chain and agricultural product sales, to provide basic data for the system; Data preprocessing module, through data cleaning to remove outliers and fill missing values, using data integration to integrate different source data formats, using feature engineering to extract valuable features for analysis and model training; Yield prediction module, for different crops, using the TransformerXL model, trained according to a large number of related sample data to obtain the predicted yield of the crops; The supply chain optimization module predicts agricultural material demand and reduces inventory accumulation by means of a mixed integer programming algorithm, optimizes logistics distribution paths by means of a deep reinforcement learning algorithm, and coordinates production and sales by means of a multi-agent reinforcement learning algorithm. The agricultural product quality traceability module uses blockchain and Internet of Things, blockchain and smart contract, and blockchain and two-dimensional code technology in the production, processing and circulation links to realize whole-process traceability of agricultural product quality. The decision support module provides decision support for enterprises in production factor allocation, industry chain coordination and market response based on the data and analysis results processed by each module, and improves comprehensive benefits.

[0015] The present application has the following advantages: Production factor allocation optimization: The present application integrates multiple heterogeneous data sources through the data acquisition layer to build an agricultural big data lake, and uses an improved federated learning framework to realize cross-farm data collaborative analysis while ensuring data privacy, effectively improving model training efficiency and optimizing production factor allocation.

[0016] Industry chain coordination enhancement: The use of federated learning and other technologies enables cross-farm data collaborative analysis, enhancing information flow and coordination between industry chain links, enabling enterprises to better respond to market changes. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flowchart of an enterprise ERP comprehensive management method based on big data according to the present application; Figure 2 is an example diagram of satellite remote sensing data according to the present application; Figure 3 is an example diagram of soil sensor data according to the present application; Figure 4 is an example diagram of unmanned aerial vehicle multispectral imaging data according to the present application; Figure 5 is an example diagram of weather data according to the present application; Figure 6 is an example diagram of agricultural material supply chain data according to the present application; Figure 7 is an example diagram of agricultural product sales data according to the present application. DETAILED DESCRIPTION

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a comprehensive enterprise ERP management method based on big data, such as... Figure 1 As shown, it includes the following steps: S01 acquires information such as land cover and crop distribution through satellite remote sensing (resolution 0.8-1.1m); it acquires parameters such as soil pH, humidity, EC value, nitrogen content, phosphorus content, potassium content, organic matter content, cation exchange capacity, available iron content, available manganese content, available copper content, available zinc content, available boron content, available molybdenum content, and total water-soluble salts through soil sensors; and it acquires five bands of multispectral imaging from UAVs, obtaining information from seven heterogeneous data sources, including red, green, blue, near-infrared, and short-wave infrared bands. In one embodiment of the present invention, it is also necessary to collect agricultural input supply chain information, which records structured information such as agricultural input procurement, storage, and transportation, and updates it in real time; at the same time, agricultural product sales information is collected, which is structured data covering information such as sales volume, sales amount, and sales region, reflecting market demand and sales trends in real time; historical data is collected, the sources of which include data accumulated from various data sources over a long period of time, for purposes such as model updates and analysis, and the collection method is to periodically archive and store data for each time period.

[0020] S02 employs an improved federated learning framework (FL-Edge) to achieve collaborative data analysis across farms while ensuring data privacy. In one embodiment of the present invention, the improved federated learning framework performs the following steps: s101, each farm (participant) uses its own data to train the model locally and obtain local model parameters; s102, each farm uploads its local model parameters to the central server (or aggregation node). s103, the central server (or aggregation node) aggregates the uploaded local model parameters according to certain rules (such as weighted average) to obtain global model parameters; s104 distributes the global model parameters to each participant, who then uses these parameters to continue the next round of local training.

[0021] In one embodiment of the present invention, the objective function of federated learning is: wherein denotes the parameter vector to be optimized; denotes the number of participants participating in federated learning; is the local objective function of the th participant, is the weight coefficient.

[0022] S03, constructing an agricultural knowledge base; In an embodiment of the present application, the specific steps of constructing the agricultural knowledge base are as follows: s201, introducing dynamic knowledge graph technology to provide technical support for constructing the knowledge base; s202, determining that the knowledge base contains three core modules, namely crop growth model (CGM), supply chain relationship network, and market prediction factor; s203, based on the above technology and modules, completing the construction of the agricultural knowledge base. The knowledge base supports semantic-level data correlation query, and the query response time is better than that of traditional SQL query.

[0023] S04, constructing a yield prediction model based on TransformerXL; the yield prediction model includes an input layer, a meteorological time series attention layer, a hidden layer, and an output layer; In an embodiment of the present application, the input layer of the yield prediction model includes: taking meteorological information closely related to yield prediction, such as temperature, precipitation, and light duration, as input. These meteorological information is organized in the form of time series, for example, recorded at daily, weekly, or monthly time intervals, to provide basic data support for subsequent layer processing, so that the model can capture the potential impact of changes in meteorological factors over time on yield.

[0024] In an embodiment of the present application, the meteorological time series attention layer of the yield prediction model includes: introducing a meteorological time series attention mechanism to process time series meteorological information, so that the model can pay more attention to the influence degree of meteorological factors at different time points on yield during processing. The core lies in the calculation of attention weights, and the calculation formula is: wherein, , are the query matrix and the key matrix, respectively, is the dimension of the key matrix , the matrix can be composed of meteorological information features at the current time step and several time steps before and after, the matrix covers meteorological information features of the entire time series. The attention weights ​This mechanism can measure the importance of meteorological information at different time steps, enabling the model to focus on key meteorological information. For example, meteorological factors during critical crop growth periods have a greater impact on yield, and the model can use this mechanism to highlight meteorological information during these periods.

[0025] In one embodiment of the present invention, the hidden layer of the yield prediction model includes further feature extraction and transformation of the data processed by the attention layer. In an agricultural context, the hidden layer uncovers the potential complex relationship between meteorological information and yield, such as the synergistic effect of different combinations of meteorological factors on yield, providing more representative features for yield prediction. These features will serve as an important basis for subsequent yield prediction.

[0026] In one embodiment of the present invention, the output layer of the yield prediction model includes: outputting the final crop yield prediction result based on the features extracted from the previous layers.

[0027] S05, to further optimize the enterprise ERP integrated management system based on big data and improve supply chain management efficiency, adopts a supply chain optimization algorithm combining mixed integer programming and deep Q-network (DQN). In one embodiment of the present invention, the key steps of the supply chain optimization algorithm are as follows: S301, Define the state and action space: Identify the state variables in the supply chain system, such as inventory levels, order quantities, and logistics transportation status, to form the state space. Simultaneously, define the actions that the system can execute, such as replenishment decisions and transportation route selection, forming the action space. ; s302, Q-value initialization function: for state-action pairs Q-value function This function performs initialization and evaluates the state. Next action Long-term cumulative rewards; s303, Perform an action and receive a reward: In state Next action The observation system transitions to a new state. And receive instant rewards. The rewards should be set in accordance with supply chain optimization goals, such as reducing logistics costs and improving service levels; s304, Update Q-value: Update the Q-value function using the following Q-value update formula: in The Q-value represents the current state and action pair; It is to perform an action The instant reward obtained afterward; The learning rate controls the step size for each update; Discount factor; The next state The maximum Q value for all possible actions; this formula combines immediate rewards. Maximum Q-value of future states To adjust the Q value of the current state-action pair.

[0028] S06 uses a lightweight consensus algorithm (PoET) to build a blockchain-enabled agricultural product quality traceability system; In one embodiment of the present invention, the basic steps of the lightweight consensus algorithm (PoET) are as follows: S401, Node participation in consensus: Each node participates in the consensus process of the agricultural product quality traceability system, laying the foundation for subsequent transaction processing and information traceability; s402, Random Selection of Validators: The PoET algorithm uses a random mechanism to select validators from participating nodes to process transactions. In this system, this random selection mechanism ensures the fairness and decentralization of the system, giving each node the opportunity to participate in the transaction verification process. S403, Transaction Processing and Consensus Reaching: The selected validator processes the transaction and reaches consensus with other nodes. This lightweight consensus algorithm ensures the efficient operation of the agricultural product quality traceability system; assuming the efficiency formula for validator transaction processing in this process is... ,in This indicates processing efficiency (the number of transactions processed per unit of time). This represents the total number of transactions. The number of validators.

[0029] In one embodiment of the present invention, an enhanced crop lifecycle management process is provided to address the sensor network reliability barriers that the system may face in extreme environments; the process includes the following steps: A low-temperature self-heating sensor packaging technology was developed. This technology employs a pulsed heating strategy, and the specific steps are as follows: Real-time monitoring of sensor temperature ; Based on the monitored temperature values, the heating power is determined according to the following rules. : when At that time, heating power ;when At that time, heating power ;when Heating power at other temperature ranges Expressed as a formula: wherein, represents the real-time temperature of the sensor; represents the heating power of the sensor; At the same time, in order to improve the accuracy of sensor data, a data compensation algorithm is introduced. The detailed steps of this algorithm are as follows: For each sensor , determine the other sensors within its neighborhood ; Calculate the weight between the sensors within the neighborhood , the calculation formula is: ; wherein and represent the positions of sensors and , respectively, is a constant parameter; Obtain the measurement values of each sensor within the neighborhood ; According to the weight and measurement value of the sensors within the neighborhood, as well as the temperature change of the sensor itself , calculate the compensated data , the calculation formula is: wherein represents the sensor number of the current data that needs to be compensated; represents the number of other sensors within the neighborhood; represents the neighborhood of the sensor ; represents the weight between the sensor and ; represents the measurement value of the sensor within the neighborhood, is the temperature compensation coefficient, ; represents the temperature change of the sensor itself.

[0030] In an embodiment of the present application, in order to enhance the crop life cycle management process, after developing the low-temperature self-heating sensor packaging technology and the data compensation algorithm, the following key steps are taken: Adopt federated learning to realize cross-base knowledge migration and build cross-domain adapters. In the case of differences in data at different bases, in order to realize effective migration of knowledge between different bases, cross-domain adapters are built through federated learning. The core formula is: where, denotes the model parameters of the target domain, denotes the model parameters of the source domain, is the learning rate, used to control the step size of parameter update. represents the source domain data and the target domain data , the distance is used to measure the difference between the two domain data distributions. The calculation formula of MMD distance is: where, and are the sample sizes of the source domain and the target domain data, , and are kernel functions, used to calculate the similarity between different data points.

[0031] An online learning mechanism is introduced to realize dynamic model updating. Considering the dynamic changes of data in actual production environment, in order to enable the model to adapt to new data in time, an online learning mechanism is introduced. The model parameter update formula is: where, is the model parameter at time t, is the model parameter at the current time, is the loss function about the gradient of the model parameter , which reflects the influence direction and degree of the current sample on the model parameter update. is the regularization term, used to prevent model overfitting, is the regularization coefficient. is the adaptive learning rate, its calculation formula is: where, is the initial learning rate, is the gradient cumulative square sum up to the current time , in this way, the learning rate will be dynamically adjusted during the training process, so that the model can quickly converge in the early stage of training, and be more stable in the later stage. A concept drift detection module is designed to trigger model updating when drift is detected. In actual application, the distribution of data may change over time, that is, the concept drift phenomenon occurs. In order to discover and respond to this situation in time, a concept drift detection module is designed. The trigger condition of drift detection is:

[0032] ​ where, is an indicator function, which is 1 when the condition in the parentheses is satisfied, indicating that a drift is detected; otherwise, it is 0. is the average value of the loss function over a time window from to time, and are the mean and standard deviation of the average value of the loss function, respectively. When the average value exceeds the mean plus three times the standard deviation, it is considered that a concept drift has occurred. Once a drift is detected, the model will be updated, and the update formula is: where, is the updated model, is the model before updating, is a drift adaptation function that adjusts the model based on the current data and historical data . The calculation formula of the drift adaptation function is: Here, is a kernel function that calculates the similarity between the current data point and historical data , and are the sample sizes of the current data and historical data, respectively. In this way, the model can adaptively update the current and historical data when a concept drift is detected to maintain the accuracy and effectiveness of the model.

[0033] In one embodiment of the present application, a specific example of the aforementioned enterprise ERP comprehensive management method based on big data is provided: This example focuses on the agricultural field and aims to address issues such as low efficiency of factor allocation, insufficient industry chain collaboration, and delayed market response. By introducing advanced technologies such as multi-source heterogeneous big data fusion, federated learning, and knowledge graph, a set of enterprise ERP comprehensive management system based on big data is constructed; including the following steps: Data preparation: In the process of constructing the enterprise ERP comprehensive management system based on big data, data preparation is a crucial step. The system collects data from multiple different data sources, which cover various aspects of agricultural production, as follows: As shown in Figure 2 , satellite remote sensing data records satellite remote sensing information collected at different times, including collection time, resolution, band number, and coverage area; As Figure 3 shown, the soil sensor data contains soil-related data collected at different times, such as collection time, soil moisture, soil temperature, soil pH, and soil conductivity; As Figure 4 shown, the unmanned aerial vehicle multispectral imaging data records the multispectral imaging information collected by the unmanned aerial vehicle, including collection time, resolution, number of bands, and coverage area; As Figure 5 shown, the weather information records weather information at different times, including collection time, temperature, humidity, wind speed, and precipitation; As Figure 6 shown, the agricultural supply chain information records agricultural procurement-related information, including agricultural product name, supplier, procurement date, procurement quantity, unit price, and total amount; As Figure 7 shown, the agricultural product sales information records agricultural product sales-related information, including agricultural product name, sales date, sales quantity, unit price, total amount, and sales region.

[0034] Data preprocessing: Data cleaning: Through the methods of outlier detection and missing value filling, the original data is cleaned to ensure the accuracy and integrity of the data; Data integration: Using data format conversion and data fusion methods, integrate data from different sources into a unified format; Feature engineering: With feature extraction and feature selection, valuable features for subsequent analysis and model training are extracted from the original data; Model prediction: For different crop types, the TransformerXL model is used, and a large number of samples are used for training, as follows: Corn: Collect climate data, soil data, planting density data, and historical yield data of corn planting areas for many years; Through deep mining and analysis of these data, the model learns the complex relationship between corn yield and various influencing factors, and finally outputs the result of corn yield prediction; Wheat: Obtain growth cycle data of different varieties of wheat, light duration data of different regions, fertilizer application data, and corresponding yield data, etc. The model fully captures the feature information in these data during training, and finally outputs the result of wheat yield prediction; Soybean: Obtain altitude data, disease and pest occurrence frequency data, irrigation quantity data, and actual yield data of soybean planting, etc.; After the model learns and optimizes the massive data, the result of soybean yield prediction is obtained; Supply chain optimization: Inventory management: Through mixed integer programming algorithm, analyze past agricultural procurement data, predict the demand of various types of agricultural materials in different seasons, effectively reduce the cost of inventory backlog; Logistics distribution: Use deep reinforcement learning algorithm, combined with real-time traffic and distribution route data, optimize the distribution path; Production and sales coordination: With the help of multi-agent reinforcement learning algorithm, integrate order data from the sales end and production capacity data from the production end, avoid overproduction or insufficient production; Quality traceability of agricultural products: Production link: Use blockchain and Internet of Things technology, install sensors in the greenhouse to collect environmental data in real time, and use blockchain technology to ensure data cannot be tampered with, accurately record the whole process data of vegetable growth; Processing link: Use blockchain, Internet of Things, smart contract and two-dimensional code technology to strictly regulate the processing process and improve the product quality pass rate; Distribution link: Use blockchain and two-dimensional code technology, consumers can obtain complete circulation information of vegetables from the production area to the sales point after scanning the code.

[0035] The above describes the embodiments of the present application, but the embodiments are not limited to the specific implementation described above, which is only illustrative and not limiting, and those skilled in the art can make more forms of equivalent embodiments under the inspiration of the embodiments, which are all within the protection of the embodiments.

Claims

1. A comprehensive enterprise ERP management method based on big data, characterized in that, Includes the following steps: Information on land cover and crop distribution can be obtained through satellite remote sensing; soil parameter information can be obtained through soil sensors; red, green, blue, near-infrared, and short-wave infrared information can be obtained through multispectral imaging by UAVs; and agricultural input supply chain data, agricultural product sales data, historical comprehensive data, and meteorological information can be obtained. The collected data is cleaned, integrated, and feature-engineered to eliminate outliers, fill in missing values, standardize the data format, and extract key features. Based on the preprocessed data, the TransformerXL deep learning model is used to predict the yield of different crops. By comprehensively utilizing mixed integer programming, deep reinforcement learning, and multi-agent reinforcement learning optimization algorithms, intelligent optimization can be achieved for enterprise inventory management, logistics and distribution, and production and sales collaborative supply chain links. By leveraging blockchain technology, we can build a quality traceability system for agricultural products that covers the entire process of production, processing, and distribution.

2. The enterprise ERP integrated management method based on big data according to claim 1, characterized in that, In the data acquisition process, the data acquisition frequencies for each data source are as follows: satellite remote sensing data: high-frequency acquisition; soil sensor data: high-frequency acquisition; UAV multispectral imaging data: periodic acquisition; meteorological information: high-frequency acquisition; agricultural input supply chain information: real-time acquisition; agricultural product sales information: real-time acquisition.

3. The enterprise ERP integrated management method based on big data according to claim 1, characterized in that, Data preprocessing steps include: Data cleaning: Outlier detection and missing value imputation methods are used to improve data quality; Data integration: Converting data from heterogeneous data sources into a unified format and merging the data; Feature engineering: Extracting key features from raw data and performing feature selection.

4. The enterprise ERP integrated management method based on big data according to claim 1, characterized in that, The yield forecasting step uses the TransformerXL model, which accurately predicts crop yields based on massive historical data and meteorological and soil influencing factors.

5. The enterprise ERP integrated management method based on big data according to claim 1, characterized in that, TransformerXL models include: Input layer: Receives time-series meteorological information and data on factors affecting yield; Meteorological Time Series Attention Layer: Focuses on the impact of meteorological factors on yield at different time steps through an attention mechanism; Hidden layer: Further extract and transform features to uncover the complex relationship between meteorology and yield; Output layer: Outputs crop yield prediction results.

6. The enterprise ERP integrated management method based on big data according to claim 1, characterized in that, The supply chain optimization steps employ one or more of the following optimization algorithms: Mixed integer programming: used for inventory management optimization to improve inventory turnover rate; Deep reinforcement learning: used for logistics and distribution optimization to improve delivery efficiency; Multi-agent reinforcement learning: used for production and sales coordination optimization to improve the accuracy of demand forecasting.

7. The enterprise ERP integrated management method based on big data according to claim 6, characterized in that, By combining mixed-integer programming with deep Q-networks, supply chain optimization is achieved through the following steps: Define the state and action space; Q-initialization function; Perform actions and receive rewards; The Q-value function is updated using the Q-value update formula.

8. The enterprise ERP integrated management method based on big data according to claim 1, characterized in that, The agricultural product quality traceability process utilizes the following key technologies: Blockchain and the Internet of Things: Used in the production process to record agricultural operation data on the blockchain, ensuring the authenticity and reliability of the data; Blockchain and smart contracts: used in the processing stage to achieve quality certification and improve product qualification rate; Blockchain and QR codes: used in the circulation process to enable product information traceability and query, shortening response time.

9. The enterprise ERP integrated management method based on big data according to claim 8, characterized in that, The blockchain uses the lightweight consensus algorithm PoET, which achieves efficient consensus through the following steps: Each node participates in consensus; Verifiers are selected randomly; Validators process transactions and reach consensus with other nodes.

10. A big data-based enterprise ERP integrated management system, used to execute the big data-based enterprise ERP integrated management method according to any one of claims 1-9, wherein the big data-based enterprise ERP integrated management system comprises the following modules: The data acquisition module is used to collect data from multiple sources, including satellite remote sensing, soil sensors, UAV multispectral imaging, meteorology, agricultural input supply chain, and agricultural product sales, to provide basic data for the system. The data preprocessing module removes outliers and fills in missing values ​​through data cleaning, integrates data from different sources using data integration, and extracts valuable features for analysis and model training using feature engineering. The yield prediction module uses the TransformerXL model for different crops, trained on a large amount of relevant sample data, to obtain the predicted crop yield. The supply chain optimization module uses mixed integer programming algorithm to predict agricultural input demand and reduce inventory backlog, uses deep reinforcement learning algorithm to optimize logistics and distribution routes, and uses multi-agent reinforcement learning algorithm to coordinate production and sales. The agricultural product quality traceability module utilizes blockchain and IoT, blockchain and smart contracts, and blockchain and QR code technologies in the production, processing, and distribution stages to achieve full traceability of agricultural product quality. The decision support module, based on the data and analysis results processed by each module, provides decision support for enterprises in terms of production factor allocation, supply chain collaboration, and market response, thereby improving overall efficiency.

Images