Multi-source data-based crop growth dynamic optimization decision-making method, system, equipment and medium

Abstract

The invention relates to the technical field of agricultural intelligent management, in particular to a crop growth dynamic optimization decision-making method, system and device based on multi-source data and a medium, and the method comprises the steps: collecting agricultural-related multi-source data; the collected data are preprocessed; loading the pre-training model; configuring hyper-parameters of the model, and introducing a low-rank matrix into a weight matrix of the pre-training model to realize parameter re-parameterization; calculating the gradient of model parameters according to the loss value during model training, and updating the parameters of the model according to the gradient; the performance of the model is evaluated by adopting an automatic evaluation index, hyper-parameters, architecture or data of the model are adjusted according to an evaluation result, and a trained prediction model, namely an agricultural agent, is generated; the agricultural agent is deployed in the agricultural decision system, so that the agricultural agent receives and processes external real-time data in real time, and provides crop growth optimization decision service according to a prediction result. And reasonable decision suggestions are provided for agricultural production.

Description

Technical Field

[0001] The present application relates to the field of intelligent agricultural management technology, and specifically to a method, system, equipment and medium for dynamic optimization decision-making of crop growth based on multi-source data. Background Art

[0002] In agricultural production, crop growth is influenced by a variety of factors. Accurate and efficient planting decisions are crucial for improving crop yield and quality while reducing resource waste. Traditional agricultural decision-making relies primarily on farmers' experience and judgment, a method that lacks scientific evidence and struggles to adapt to the complex and changing agricultural environment, leading to low production efficiency and inefficient resource utilization.

[0003] With the development of information technology, the application of multi-source data in agriculture has gradually gained momentum. Meteorological data can reflect the impact of weather changes on crop growth, data from IoT devices can monitor soil conditions in real time, and crop growth data and historical agricultural data contain information about crop growth patterns and past planting experience. However, this data is scattered and in various formats, making the effective integration and utilization of these data a key challenge in the development of intelligent agriculture.

[0004] Furthermore, in terms of model application, directly using complex deep learning models for training often faces challenges such as high computing resource requirements, long training times, and overfitting. Existing crop growth decision-making models suffer from flaws in training and optimization, making it difficult to accurately predict crop growth and provide reliable decision-making recommendations. Summary of the Invention

[0005] In response to the problem that existing crop growth decision models have defects in model training and optimization, making it difficult to accurately predict crop growth and provide reliable decision-making recommendations, the present invention provides a crop growth dynamic optimization decision method, system, equipment and medium based on multi-source data.

[0006] In a first aspect, the technical solution of the present invention provides a method for dynamic optimization decision-making of crop growth based on multi-source data, comprising the following steps: Collect multi-source data related to agriculture, including meteorological data, IoT device data, crop growth data, and historical agricultural data; Preprocess the collected data and integrate multi-source data according to time and location dimensions to form a unified data set; divide the data set into training set, validation set and test set; Load the pre-trained model and design the input layer, hidden layer, and output layer; Configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization; Input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient; Quantitative analysis of model performance using automatic evaluation metrics based on the validation set; Based on the test set and the evaluation results, the model's hyperparameters, architecture, or data are adjusted to generate a trained prediction model, i.e., an agricultural agent. Deploy the agricultural intelligent agent into the agricultural decision-making system, so that the agricultural intelligent agent can receive and process external real-time data in real time, and provide crop growth optimization decision-making services based on the prediction results.

[0007] As a further limitation of the technical solution of the present invention, the collected data is preprocessed, and the multi-source data is integrated according to time and location dimensions to form a unified data set; and the step of dividing the data set into a training set, a validation set, and a test set includes: The collected data is cleaned, samples are labeled for tasks that require supervised learning, features are extracted and transformed, and multi-source data is integrated according to time and location dimensions to form a unified data set, which is divided into training set, validation set, and test set.

[0008] As a further limitation of the technical solution of the present invention, the steps of loading the pre-trained model and designing the input layer, hidden layer, and output layer include: Select and load a pre-trained model; Design the input layer and design corresponding input modules according to different types of data; Design hidden layers for feature extraction and representation learning to capture complex patterns and relationships in the data; Design the output layer and design different output modules according to different tasks.

[0009] As a further limitation of the technical solution of the present invention, the steps of configuring the hyperparameters of the model, selecting the loss function, and introducing a low-rank matrix into the weight matrix of the pre-trained model to implement parameter reparameterization include: Set the initial learning rate to control the update step size of model parameters; Set the number of samples input to the model during each training; Set the number of times the model is trained on the entire training set; Choose a loss function based on the task type; Use the low-rank adaptive fine-tuning method to introduce a low-rank matrix into the weight matrix of the pre-trained model to achieve parameter reparameterization; The low-rank matrix decomposition formula is as follows:

[0010] in, are the initial parameters of the pre-trained model, are the parameters that need to be updated. and are two low-rank matrices.

[0011] As a further limitation of the technical solution of the present invention, the steps of inputting samples in the training set into the model, calculating the loss value using the selected loss function, calculating the gradient of the model parameters according to the loss value, and updating the parameters of the model according to the gradient include: Input the samples in the training set into the model, and obtain the predicted output through the model calculation; Calculate the loss value using the selected loss function based on the predicted output and the true label; Calculate the gradient of the model parameters based on the loss value, and propagate the gradient to each parameter of the model through the back propagation algorithm; Use the optimizer to update the model parameters according to the gradient, and repeat the model training steps until the specified number of training rounds is reached or the stopping condition is met.

[0012] As a further limitation of the technical solution of the present invention, providing a crop growth optimization decision service based on the prediction results of the model includes: The prediction model generates recommendations for farming activities based on set rules and strategies; If the soil fertility assessment indicates a nutrient deficiency, the model calculates and recommends the type and amount of fertilizer required based on the nutrient requirements of the crop variety and growth stage; When soil moisture is below the suitable range for crop growth, the model refers to meteorological data to determine the time and amount of watering; Combined with crop disease and pest monitoring data and current meteorological conditions, the possibility and severity of disease and pest occurrence are judged, and the type of pesticide, application time and spraying dosage are recommended.

[0013] In terms of soil fertilizer testing, the technical solution of the present invention can accurately recommend fertilizer types and amounts based on soil nutrient status, crop varieties, and growth stages; in terms of watering decisions, it can comprehensively determine the appropriate watering time and amount by combining soil moisture and meteorological data; in terms of pest and disease control, it can accurately determine the occurrence of pests and diseases and recommend appropriate pesticides and drug use plans based on pest and disease monitoring data and meteorological conditions, effectively improving the accuracy and efficiency of agricultural production.

[0014] Presenting soil fertility testing, crop suitability assessment results, and agricultural activity recommendations to users through a visual interface greatly improves the intuitiveness and accessibility of decision-making information. Farmers can easily understand and implement these recommendations based on their farming practices without requiring specialized data analysis and modeling knowledge, lowering the barrier to entry for intelligent agricultural decision-making.

[0015] As a further limitation of the technical solution of the present invention, the method further includes: Based on the agricultural decision-making system, soil fertility testing, crop suitability assessment results and agricultural activity recommendations are presented to users through a visual interface.

[0016] In a second aspect, the technical solution of the present invention also provides a crop growth dynamic optimization decision system based on multi-source data, including a data collection module, a data preprocessing module, an architecture design module, a setting module, a training module, a verification module, a testing module and a deployment module; Data collection module, used to collect multi-source data related to agriculture, including meteorological data, IoT device data, crop growth data, and historical agricultural data; The data preprocessing module is used to preprocess the collected data and integrate multi-source data according to time and location dimensions to form a unified data set; the data set is divided into training set, validation set and test set; Architecture design module, used to load pre-trained models and design input layers, hidden layers, and output layers; The setup module is used to configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization; For the soil fertility measurement task, the mean square error loss function is used to measure the difference between the fertilizer application amount predicted by the model and the actual application amount. For the crop suitability assessment task, the cross entropy loss function is used to measure the difference between the crop suitability category probability predicted by the model and the true label. The training module is used to input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient; The validation module is used to quantitatively analyze the model performance using automatic evaluation indicators based on the validation set; The testing module is used to adjust the model's hyperparameters, architecture, or data based on the evaluation results of the test set to generate a trained prediction model, namely the agricultural agent; The deployment module is used to deploy the agricultural intelligent agent into the agricultural decision-making system, enabling the agricultural intelligent agent to receive and process external real-time data in real time and provide crop growth optimization decision-making services based on the prediction results.

[0017] Crop growth optimization decision-making services provided based on the model's prediction results include: the prediction model generates agricultural activity recommendations according to set rules and strategies; if the soil fertility assessment shows a lack of certain nutrients, the model calculates and recommends the appropriate fertilizer type and amount based on the nutritional needs of the crop variety and growth stage; if the soil moisture is below the suitable range for crop growth, the model refers to meteorological data to determine the time and amount of watering; combined with crop disease and pest monitoring data and current meteorological conditions, the model determines the possibility and severity of disease and pest occurrence and recommends the type of pesticide, application time and spraying dosage.

[0018] In a third aspect, the technical solution of the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the dynamic optimization decision method for crop growth based on multi-source data as described in the first aspect.

[0019] In a fourth aspect, the technical solution of the present invention also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable the computer to execute the dynamic optimization decision method for crop growth based on multi-source data as described in the first aspect.

[0020] It can be seen from the above technical solutions that the present application has the following advantages: comprehensive acquisition of various types of information affecting crop growth. By integrating these data into a unified data set for model training, compared with a single data source, it is possible to explore richer crop growth patterns and relationships between influencing factors, providing a solid data foundation for accurate decision-making. A pre-trained model is adopted and combined with the low-rank adaptive fine-tuning method (LoRA), and a low-rank matrix is introduced into the pre-trained model weight matrix to achieve parameter re-parameterization. The computational complexity and resource consumption of model training are significantly reduced. With limited computing resources, the model can more efficiently learn specific knowledge in the agricultural field, while reducing the risk of overfitting and improving the model's generalization ability and prediction accuracy. After being trained and optimized with multi-source data, the agricultural intelligent agent can be deployed to the agricultural decision-making system to process external real-time data in real time. Based on the model prediction results, crop growth optimization decision-making services can be accurately provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A flowchart of a method provided in an embodiment of the present invention.

[0023] Figure 2 A connection block diagram of the system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, features, and advantages of this application more obvious and easy to understand, the technical solutions protected by this application will be clearly and completely described below using specific embodiments and drawings. Obviously, the embodiments described below are only part of the embodiments of this application, not all of them. Based on the embodiments in this patent, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this patent.

[0025] like Figure 1 As shown, an embodiment of the present invention provides a method for dynamic optimization decision-making of crop growth based on multi-source data, comprising the following steps: S1: Collect multi-source data related to agriculture, including meteorological data, IoT device data, crop growth data, and historical agricultural data; In this step, meteorological data includes temperature, humidity, sunshine duration, rainfall, etc. IoT device data includes soil moisture sensor data and soil fertility sensor data. Crop growth data includes crop variety, growth stage, plant height, leaf area, etc. Historical agricultural data includes agricultural production records and pest and disease occurrences from past years.

[0026] S2: Preprocess the collected data and integrate the multi-source data according to the time and location dimensions to form a unified dataset; divide the dataset into training set, validation set and test set; The preprocessing process in this step specifically includes: S21: Data cleaning removes noise, outliers, and missing values from the data. For missing values, methods such as mean filling and interpolation can be used to handle them.

[0027] S22: For tasks that require supervised learning, label the samples accordingly.

[0028] It should be noted that labeling samples for different supervised learning tasks allows the model to learn the mapping relationship between input data and expected output.

[0029] The following are examples of labeling for different task scenarios: 1. The soil fertility assessment task is used to predict the content of various nutrients (such as nitrogen, phosphorus, and potassium) in the soil and recommend appropriate fertilizer types and amounts based on the content. IoT devices are used to collect soil sample data such as soil pH, electrical conductivity, and moisture, as well as historical fertilization records and crop growth data. The nitrogen, phosphorus, and potassium content of the soil sample is used as the label. For example, if a soil sample contains 120 mg / kg of nitrogen, 30 mg / kg of phosphorus, and 150 mg / kg of potassium, the label for the sample is [120, 30, 150].

[0030] 2. The crop pest and disease prediction task combines crop pest and disease monitoring data with meteorological conditions to determine the likelihood and severity of pest and disease occurrence and recommend pesticide types, application times, and spraying dosages. Crop image data, meteorological data (temperature, humidity, rainfall, etc.), and pest and disease monitoring data (such as pest populations and disease symptoms) are collected. For binary classification problems: If only determining whether a pest and disease has occurred, label it as 1 if it has occurred and 0 if it has not. For multi-classification problems: If determining the type of pest and disease, such as powdery mildew, aphids, or spider mites, label them with the corresponding category number, e.g., powdery mildew is labeled 1, aphids are labeled 2, and spider mites are labeled 3. Severity grading: If determining the severity of a pest and disease, it can be categorized as mild, moderate, or severe, labeled 1, 2, and 3, respectively.

[0031] 3. The Crop Growth Stage Prediction task predicts crop growth stages, such as seedling, flowering, and fruiting, based on crop growth data (plant height, leaf area, stem diameter, etc.) and meteorological data. Crop growth indicators are regularly measured, and meteorological data is recorded. Each growth stage is assigned a unique number, such as 1 for seedling stage, 2 for flowering stage, and 3 for fruiting stage.

[0032] 4. The irrigation decision task integrates soil moisture and meteorological data to determine the appropriate watering time and amount. Soil moisture sensors collect soil moisture data, along with meteorological data (temperature, humidity, light intensity, etc.). Watering is marked as 1 if necessary, and 0 if not. The specific amount of water (liters) is used as the label. For example, if 5 liters of water are needed at a time, the label is 5.

[0033] S23: Extract features from data, such as extracting seasonal features from meteorological data; transform data, such as normalization and standardization, to make the data comparable.

[0034] S3: Load the pre-trained model and design the input layer, hidden layer, and output layer; In this step, the DeepSeek-R1 pre-trained model is loaded, and the model architecture design includes the following: Input layer design: Design corresponding input modules based on different types of data (such as meteorological data, IoT device data, etc.) so that the data can be correctly input into the model.

[0035] Hidden layer design: Design hidden layers for feature extraction and representation learning, capturing complex patterns and relationships in the data through multiple hidden layers.

[0036] Output layer design: Design different output modules according to different tasks (such as soil fertility prediction, pest and disease occurrence possibility prediction, etc.).

[0037] S4: Configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization. This step specifically includes: S41: Set the initial learning rate to control the update step size of the model parameters, for example, set it to 0.001. Set the number of samples input to the model during each training, that is, the batch size, for example, set it to 32. Set the number of times the model trains the entire training set, that is, the number of training rounds, for example, set it to 100 rounds.

[0038] S42: Select loss function according to task type; In this step, the soil fertilizer measurement task can use the mean square error loss function (MSE) to measure the difference between the fertilizer usage predicted by the model and the actual usage.

[0039] The crop suitability assessment task uses the cross-entropy loss function to measure the difference between the crop suitability class probability predicted by the model and the true label.

[0040] The agricultural activity recommendation task can select an appropriate loss function according to the specific task type, such as using the cross entropy loss function for classification tasks and the mean square error loss function for regression tasks.

[0041] S43: Use the low-rank adaptive fine-tuning method to introduce a low-rank matrix into the weight matrix of the pre-trained model to achieve parameter reparameterization; The low-rank matrix decomposition formula is as follows:

[0042] in, are the initial parameters of the pre-trained model, are the parameters that need to be updated. and are two low-rank matrices.

[0043] During the model training process, the low-rank matrix improves parameter efficiency by reparameterization strategy, reducing the number of trainable parameters, and reducing computational complexity.

[0044] Assume that the weight matrix of the pre-trained model is W Dimension 0 is d × k If you train directly W 0, then d × k parameters. By using the low-rank adaptive fine-tuning method, the low-rank matrix introduced B The dimension is d × r , A The dimension is r ×k ( r is the rank of the low-rank matrix, r ≪ d , k ), the number of trainable parameters is only d × r + r × k For example, when d =1000, k =1000, r =10, directly training the weight matrix has 1000×1000=1000000 parameters, while using low-rank matrix training only has 1000×10+10×1000=20000 parameters. The number of trainable parameters is significantly reduced, and training can be completed more efficiently with limited computing resources.

[0045] It should be noted that d Represents the number of rows in the weight matrix. In the context of neural networks, for the weight matrix of a certain layer, d Usually corresponds to the dimension of the layer output, that is, the number of neurons in the layer. For example, in a fully connected layer, if the layer has d neurons, then the weight matrix in the output direction has d elements. k Represents the number of columns in the weight matrix. In a neural network, k Generally corresponds to the dimension of the input of this layer, that is, the number of neurons in the previous layer or the number of input features. r The core idea of low-rank matrix decomposition is to approximate a high-dimensional weight matrix into two low-rank matrices B and A The product of r Usually much smaller than d and k By introducing low-rank matrices, the number of parameters that need to be trained can be reduced, thereby improving training efficiency.

[0046] S5: Input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient; This step specifically includes: S51: Input the samples in the training set into the model, and obtain the predicted output through the model calculation; S52: Calculate the loss value using the selected loss function based on the predicted output and the true label; S53: Calculate the gradient of the model parameters based on the loss value, and propagate the gradient to each parameter of the model through the back propagation algorithm; S54: Use the optimizer to update the model parameters according to the gradient, and repeat the model training steps until the specified number of training rounds is reached or the stopping condition is met.

[0047] During the forward and back propagation of model training, the amount of computation is closely related to the number of parameters. Since the low-rank matrix reduces the number of trainable parameters, the amount of computation required to calculate the model output during forward propagation is reduced accordingly. When calculating the gradient during back propagation, the computational complexity is also greatly reduced. This is because only the low-rank matrix needs to be calculated. A and B The gradient of , without having to calculate the entire weight matrix W This significantly reduces the computational overhead of the model at each training iteration, improves training efficiency, and enables training to be completed in a shorter time, or more training iterations to be performed in the same amount of time, helping the model converge to a better solution faster.

[0048] S6: Quantify the model performance using automatic evaluation indicators based on the validation set. In this step, the model performance is quantitatively analyzed using automatic evaluation indicators (such as mean square error, accuracy, recall, etc.) based on the validation set to understand the model's performance on the validation set.

[0049] S7: Based on the test set and the evaluation results, the model's hyperparameters, architecture, or data are adjusted to generate a trained prediction model, i.e., an agricultural agent. Use the test dataset to thoroughly evaluate the trained model to ensure it meets expected performance standards across key metrics such as accuracy, recall, and F1 score. For agricultural agent models, additional attention may be paid to their practicality and interpretability in real-world agricultural scenarios. Based on the evaluation results, optimize the model. This may include adjusting model hyperparameters, increasing training data, and refining the model structure to further improve model performance and stability.

[0050] Based on the evaluation results of the validation and test sets, adjust the model's hyperparameters, such as the learning rate, batch size, and number of training rounds, to further improve the model's performance. If the model's performance is still unsatisfactory, consider adjusting the model's architecture, such as increasing or decreasing the number of hidden layers or adjusting the number of neurons.

[0051] It should be noted that the specific steps for hyperparameter adjustment during the training process are as follows: S71: Randomly select some hyperparameter combinations for training and evaluation to obtain initial performance data.

[0052] S72: Build a probabilistic model, such as a Gaussian process model, based on historical hyperparameter combinations and performance data.

[0053] S73: Based on the probability model, select the next hyperparameter combination that is most likely to improve model performance.

[0054] S74: Train the model using the selected hyperparameter combination and evaluate the model performance on the validation set.

[0055] S75: Add new hyperparameter combinations and performance data to historical data and update the probability model.

[0056] Repeat steps S73–S75 until a stopping condition is met, such as reaching the maximum number of iterations or the model performance no longer improves.

[0057] S8: Deploy the agricultural intelligent agent into the agricultural decision-making system, so that the agricultural intelligent agent can receive and process external real-time data in real time, and provide crop growth optimization decision-making services based on the prediction results.

[0058] Select an appropriate deployment platform based on the specific needs and operating environment of the agricultural decision-making system. Common options include cloud computing platforms, local servers, or edge computing devices. Cloud computing platforms offer powerful computing resources and flexible scalability, making them suitable for processing large-scale agricultural data and complex model inference tasks. Local servers can meet the needs of scenarios with high data privacy and real-time requirements. Edge computing devices are more suitable for real-time decision-making on-site, reducing data transmission latency.

[0059] Convert the trained model to a format suitable for the deployment platform. Identify the dependencies required for model execution, such as programming language environments, deep learning frameworks, related libraries, and tools, and install and configure them on the deployment platform. Ensure that the deployment environment is as consistent as possible with the model training environment to avoid model failures due to dependency issues.

[0060] If you choose a cloud computing platform, you can typically deploy the model using the services and tools provided by the platform. For example, you can use services such as Function Compute, Container Service, or the machine learning platform PAI to deploy the model. Upload the converted model file to the cloud platform and configure and deploy it according to the platform's documentation. This configuration process may require specifying parameters such as the model's input and output formats, computing resource allocation, and network settings. Once deployed, the cloud platform provides corresponding APIs for interacting with the model and implementing agricultural decision-making functions.

[0061] Connect the deployed model to other components of the agricultural decision-making system through data interfaces. Agricultural decision-making systems typically include data acquisition, data storage, and decision analysis modules. The model requires real-time agricultural data (such as soil moisture, meteorological data, and crop growth information) from the data acquisition module as input. After model inference, the model outputs decision results (such as irrigation recommendations, fertilization plans, and pest control measures) to the decision analysis module for reference and implementation by agricultural producers. Therefore, it is necessary to define data formats, transmission protocols, and interface specifications to ensure smooth data exchange between the model and other components.

[0062] Integrate the model's decision results with the business logic of the agricultural decision-making system. Develop appropriate decision rules and strategies based on the actual processes and needs of agricultural production. For example, by further analyzing and processing the model's output based on factors such as crop growth stage, soil conditions, and weather forecasts, specific agricultural production operation recommendations can be generated. Furthermore, the model's predictions can be compared and integrated with historical data and expert experience to improve the accuracy and reliability of decisions.

[0063] Design a user-friendly interface for agricultural producers so they can easily view the model's decision results and related information. This user interface can be a web application or mobile app that displays real-time agricultural data, model predictions, recommended agricultural measures, and more. The interface should be concise and easy to use, allowing agricultural producers to quickly understand and apply the model's decision recommendations, thereby improving the efficiency and quality of agricultural production.

[0064] Establish a model performance monitoring mechanism to monitor the model's performance indicators in real-time, such as prediction accuracy, inference speed, and resource utilization. By monitoring data, potential performance degradation, abnormal behavior, or potential problems in the model can be promptly identified. For example, if the model's predictions deviate significantly from the actual situation, or if excessive inference time affects decision-making efficiency, timely analysis and resolution are necessary. As the agricultural production environment changes and data continues to accumulate, the model may need to be updated and optimized. Regularly collect new agricultural data, retrain and evaluate the model, and adjust the model's parameters and structure based on new data and business needs to maintain model accuracy and adaptability. At the same time, promptly deploy the updated model to the actual system to ensure that the agricultural decision-making system can always make decisions based on the latest model.

[0065] In an embodiment of the present invention, providing a crop growth optimization decision service based on the prediction results of the model includes: The prediction model generates recommendations for agricultural activities based on set rules and strategies. If soil fertility assessments indicate a nutrient deficiency, the model calculates and recommends appropriate fertilizer types and amounts based on the nutritional needs of the crop variety and growth stage. If soil moisture is below the suitable range for crop growth, the model refers to meteorological data to determine the timing and amount of watering. Combined with crop pest and disease monitoring data and current meteorological conditions, the model determines the likelihood and severity of pest and disease occurrence and recommends pesticide types, application times, and spraying dosages.

[0066] In terms of soil fertilizer testing, the technical solution of the present invention can accurately recommend fertilizer types and amounts based on soil nutrient status, crop varieties, and growth stages; in terms of watering decisions, it can comprehensively determine the appropriate watering time and amount by combining soil moisture and meteorological data; in terms of pest and disease control, it can accurately determine the occurrence of pests and diseases and recommend appropriate pesticides and drug use plans based on pest and disease monitoring data and meteorological conditions, effectively improving the accuracy and efficiency of agricultural production.

[0067] Presenting soil fertility testing, crop suitability assessment results, and agricultural activity recommendations to users through a visual interface greatly improves the intuitiveness and accessibility of decision-making information. Farmers can easily understand and implement these recommendations based on their farming practices without requiring specialized data analysis and modeling knowledge, lowering the barrier to entry for intelligent agricultural decision-making.

[0068] Based on the agricultural decision-making system, soil fertility testing, crop suitability assessment results and agricultural activity recommendations are presented to users through a visual interface.

[0069] like Figure 2 As shown, an embodiment of the present invention further provides a crop growth dynamic optimization decision system based on multi-source data, including a data collection module, a data preprocessing module, an architecture design module, a setting module, a training module, a verification module, a testing module and a deployment module; Data collection module, used to collect multi-source data related to agriculture, including meteorological data, IoT device data, crop growth data, and historical agricultural data; The data preprocessing module is used to preprocess the collected data and integrate multi-source data according to time and location dimensions to form a unified data set; the data set is divided into training set, validation set and test set; Architecture design module, used to load pre-trained models and design input layers, hidden layers, and output layers; The setup module is used to configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization; The training module is used to input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient; The validation module is used to quantitatively analyze the model performance using automatic evaluation indicators based on the validation set; The testing module is used to adjust the model's hyperparameters, architecture, or data based on the evaluation results of the test set to generate a trained prediction model, namely the agricultural agent; The deployment module is used to deploy the agricultural intelligent agent into the agricultural decision-making system, enabling the agricultural intelligent agent to receive and process external real-time data in real time and provide crop growth optimization decision-making services based on the prediction results.

[0070] In some embodiments, the data preprocessing module is specifically used to clean the collected data, label samples for tasks that require supervised learning, extract and transform features from the data, and integrate multi-source data according to time and location dimensions to form a unified data set, which is divided into a training set, a validation set, and a test set.

[0071] In some embodiments, the architecture design module is specifically used to select and load a pre-trained model; design the input layer to design corresponding input modules according to different types of data; design the hidden layer to perform feature extraction and representation learning to capture complex patterns and relationships in the data; and design the output layer to design different output modules according to different tasks.

[0072] In some embodiments, a setting module is specifically used to set the initial learning rate to control the update step size of the model parameters; set the number of samples input to the model during each training; set the number of times the model trains the entire training set; select a loss function based on the task type; use a low-rank adaptive fine-tuning method to introduce a low-rank matrix into the weight matrix of the pre-trained model to achieve parameter reparameterization; the low-rank matrix decomposition formula is as follows:

[0073] in, are the initial parameters of the pre-trained model, are the parameters that need to be updated. and are two low-rank matrices.

[0074] In some embodiments, the training module is specifically used to input samples in the training set into the model, and obtain the predicted output through calculation by the model; calculate the loss value using the selected loss function based on the predicted output and the true label; calculate the gradient of the model parameters based on the loss value, and propagate the gradient to each parameter of the model through the back propagation algorithm; use the optimizer to update the parameters of the model according to the gradient, and repeat the model training steps until the specified number of training rounds is reached or the stopping condition is met.

[0075] In some embodiments, providing crop growth optimization decision services based on the model's prediction results includes: the prediction model generates agricultural activity recommendations according to set rules and strategies; if the soil fertility assessment shows a lack of certain nutrients, the model calculates and recommends appropriate fertilizer types and amounts based on the nutritional requirements of the crop variety and growth stage; if the soil moisture is lower than the suitable range for crop growth, the model refers to meteorological data to determine the time and amount of watering; combined with crop disease and pest monitoring data and current meteorological conditions, the model determines the possibility and severity of disease and pest occurrence, and recommends pesticide types, application time and spraying dosage.

[0076] In some embodiments, the system further includes a visualization display module for presenting soil fertility testing, crop suitability assessment results, and agricultural activity recommendations to the user through a visualization interface based on the agricultural decision-making system.

[0077] An embodiment of the present invention also provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus. The communication bus can be used for information transmission between electronic devices and sensors. The processor can call the logic instructions in the memory to execute the following method: S1: Collect multi-source data related to agriculture, including meteorological data, Internet of Things device data, crop growth data and historical agricultural data; S2: Pre-process the collected data, and integrate the multi-source data according to the time and location dimensions to form a unified data set; divide the data set into a training set, a validation set and a test set; S3: Load the pre-trained model, design the input layer, hidden layer and output layer; S4: Configure the hyperparameters of the model, select the loss function, and introduce a low-rank matrix into the weight matrix of the pre-trained model to achieve parameter weighting. Parameterization; S5: Input the samples in the training set into the model, use the selected loss function to calculate the loss value, calculate the gradient of the model parameters according to the loss value, and update the model parameters according to the gradient; S6: Use automatic evaluation indicators based on the validation set to quantitatively analyze the model performance; S7: Based on the test set, adjust the model's hyperparameters, architecture or data according to the evaluation results to generate a trained prediction model, namely the agricultural intelligent agent; S8: Deploy the agricultural intelligent agent into the agricultural decision-making system, so that the agricultural intelligent agent can receive and process external real-time data in real time, and provide crop growth optimization decision-making services based on the prediction results.

[0078] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0079] An embodiment of the present invention provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable a computer to execute the method provided by the above method embodiment, for example, including: S1: collecting multi-source data related to agriculture, including meteorological data, Internet of Things device data, crop growth data, and historical agricultural data; S2: pre-processing the collected data and integrating the multi-source data according to time and location dimensions to form a unified data set; dividing the data set into a training set, a validation set, and a test set; S3: loading a pre-trained model, designing the input layer, hidden layer, and output layer; S4: configuring the hyperparameters of the model and selecting the loss function , and introduce a low-rank matrix into the weight matrix of the pre-trained model to realize parameter reparameterization; S5: input the samples in the training set into the model, use the selected loss function to calculate the loss value, calculate the gradient of the model parameters according to the loss value, and update the parameters of the model according to the gradient; S6: Based on the validation set, use automatic evaluation indicators to quantitatively analyze the model performance; S7: Based on the test set, adjust the hyperparameters, architecture or data of the model according to the evaluation results to generate a trained prediction model, namely the agricultural intelligent agent; S8: deploy the agricultural intelligent agent into the agricultural decision-making system, so that the agricultural intelligent agent can receive and process external real-time data in real time, and provide crop growth optimization decision-making services based on the prediction results.

[0080] The above description of the disclosed embodiments will enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in their embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A crop growth dynamic optimization decision-making method based on multi-source data, characterized in that: The following steps are involved: Collect multi-source data related to agriculture, including meteorological data, IoT device data, crop growth data, and historical agricultural data; Preprocess the collected data and integrate multi-source data according to time and location dimensions to form a unified data set; Divide the dataset into training, validation, and test sets; Load the pre-trained model and design the input layer, hidden layer, and output layer; Configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization; Input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient; Quantitative analysis of model performance using automatic evaluation metrics based on the validation set; Based on the test set and the evaluation results, the model's hyperparameters, architecture, or data are adjusted to generate a trained prediction model, i.e., an agricultural agent. Deploy the agricultural intelligent agent into the agricultural decision-making system, so that the agricultural intelligent agent can receive and process external real-time data in real time, and provide crop growth optimization decision-making services based on the prediction results.

2. The crop growth dynamic optimization decision-making method based on multi-source data according to claim 1 is characterized in that: Preprocess the collected data and integrate multi-source data according to time and location dimensions to form a unified data set; The steps to split the dataset into training, validation, and test sets include: The collected data is cleaned, samples are labeled for tasks that require supervised learning, features are extracted and transformed, and multi-source data is integrated according to time and location dimensions to form a unified data set, which is divided into training set, validation set, and test set.

3. The crop growth dynamic optimization decision-making method based on multi-source data according to claim 2 is characterized in that: The steps to load the pre-trained model and design the input layer, hidden layer, and output layer include: Select and load a pre-trained model; Design the input layer and design corresponding input modules according to different types of data; Design hidden layers for feature extraction and representation learning to capture complex patterns and relationships in the data; Design the output layer and design different output modules according to different tasks.

4. The crop growth dynamic optimization decision-making method based on multi-source data according to claim 3 is characterized in that: The steps to configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization include: Set the initial learning rate to control the update step size of model parameters; Set the number of samples to be input into the model each time training; Set the number of times the model is trained on the entire training set; Choose a loss function based on the task type; Use the low-rank adaptive fine-tuning method to introduce a low-rank matrix into the weight matrix of the pre-trained model to achieve parameter reparameterization; The low-rank matrix decomposition formula is as follows: in, are the initial parameters of the pre-trained model, are the parameters that need to be updated. and are two low-rank matrices.

5. The crop growth dynamic optimization decision-making method based on multi-source data according to claim 4 is characterized in that: Input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient. The steps include: Input the samples in the training set into the model, and obtain the predicted output through the model calculation; Calculate the loss value using the selected loss function based on the predicted output and the true label; Calculate the gradient of the model parameters based on the loss value, and propagate the gradient to each parameter of the model through the back propagation algorithm; Use the optimizer to update the model parameters according to the gradient, and repeat the model training steps until the specified number of training rounds is reached or the stopping condition is met.

6. The crop growth dynamic optimization decision-making method based on multi-source data according to claim 5, characterized in that: Crop growth optimization decision-making services based on the model's prediction results include: The prediction model generates recommendations for farming activities based on set rules and strategies; If the soil fertility assessment indicates a nutrient deficiency, the model calculates and recommends the type and amount of fertilizer required based on the nutrient requirements of the crop variety and growth stage; When soil moisture is below the suitable range for crop growth, the model refers to meteorological data to determine the time and amount of watering; Combined with crop disease and pest monitoring data and current meteorological conditions, the possibility and severity of disease and pest occurrence are judged, and the type of pesticide, application time and spraying dosage are recommended.

7. The crop growth dynamic optimization decision-making method based on multi-source data according to claim 6, characterized in that: The method further includes: Based on the agricultural decision-making system, soil fertility testing, crop suitability assessment results and agricultural activity recommendations are presented to users through a visual interface.

8. A crop growth dynamic optimization decision system based on multi-source data, characterized by: It includes data collection module, data preprocessing module, architecture design module, setup module, training module, verification module, testing module and deployment module; Data collection module, used to collect multi-source data related to agriculture, including meteorological data, IoT device data, crop growth data, and historical agricultural data; The data preprocessing module is used to preprocess the collected data and integrate multi-source data according to time and location dimensions to form a unified data set; Divide the dataset into training, validation, and test sets; Architecture design module, used to load pre-trained models and design input layers, hidden layers, and output layers; The setup module is used to configure the model's hyperparameters, select the loss function, and introduce a low-rank matrix into the pre-trained model's weight matrix to achieve parameter reparameterization; The training module is used to input the samples in the training set into the model, calculate the loss value using the selected loss function, calculate the gradient of the model parameters based on the loss value, and update the model parameters based on the gradient; The validation module is used to quantitatively analyze the model performance using automatic evaluation indicators based on the validation set; The testing module is used to adjust the model's hyperparameters, architecture, or data based on the evaluation results of the test set to generate a trained prediction model, namely the agricultural agent; The deployment module is used to deploy the agricultural intelligent agent into the agricultural decision-making system, enabling the agricultural intelligent agent to receive and process external real-time data in real time and provide crop growth optimization decision-making services based on the prediction results.

9. An electronic device, characterized in that: The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor to enable the at least one processor to execute the crop growth dynamic optimization decision-making method based on multi-source data as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, which enable the computer to execute the crop growth dynamic optimization decision-making method based on multi-source data according to any one of claims 1 to 7.

Images