Agricultural machine terminal operation real-time interaction system based on NLP large model and control method

Abstract

The invention discloses an agricultural machinery terminal operation real-time guidance and interaction system based on an NLP large model. The system comprises the steps of demand analysis, data acquisition and preparation, model selection and training, real-time data stream processing, user interface design, development of the real-time guidance and interaction system, testing and optimization and the like. According to a large amount of data, training and model verification are carried out, voice interaction between a user and an agricultural machine terminal is realized, and the user uses a verbal instruction to command an agricultural machine to realize operation instructions such as sowing, fertilization, an advancing path, operation depth and fertilization amount; the agricultural machine terminal system can monitor the operation process in real time and provide feedback to the user, including operation progress, operation quality, agricultural machine state and the like. According to the system, the operation effect can be optimized, and real-time and intelligent operation suggestions are provided for agricultural machinery operation.

Description

Technical Field

[0001] The present invention relates to a real-time interactive system and control method for agricultural machinery terminal operations, and in particular to a real-time interactive system and control method for agricultural machinery terminal operations based on an NLP large model, and belongs to the technical field of agricultural machinery. Background Art

[0002] With the development of technology, modern information technology has been deeply integrated with agricultural operations. In order to improve the intelligence of agricultural machinery and equipment, the European Machinery Manufacturers Association emphasized the role of the Internet of Things, big data, and artificial intelligence in improving the intelligence of agricultural machinery and equipment at the meeting. In particular, neural networks, algorithms and their learning in artificial intelligence play an important role in improving the intelligence of agricultural machinery and equipment. Zhao Chunjiang believes that the core of machine intelligence is computing, but algorithms alone are not enough, and data is also needed. Data relies on the Internet of Things, including communications and sensors. "In the future, we need to improve the intelligence of agricultural machinery and equipment. Zhang, Y et al., published a paper titled "A Review of Human-Machine Interaction in Agriculture Based on Natural Language Processing" in the journal "Journal of Agricultural Science and Technology" in 2022, pointing out that natural language processing (NLP) technology can help farmers interact with agricultural machinery more intelligently, thereby improving the efficiency and quality of agricultural production. Wang, Y et al., published a paper titled "Research on the Interaction between Agricultural Machinery and Farmers Based on Natural Language Processing" in the Journal of Agricultural Mechanization Research in 2023, proposing an agricultural machinery interaction method based on natural language processing, which can improve the operating efficiency of agricultural machinery and reduce the difficulty of operation. However, this method still requires a lot of human intervention in complex environments. The significance of studying agricultural machinery interaction methods based on natural language processing is to improve the efficiency of agricultural machinery operation and reduce the difficulty of operation. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a real-time guidance and interaction system and control method for agricultural machinery terminal operations based on a large NLP model.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solution:

[0005] A real-time guidance and interaction system for agricultural machinery terminal operations based on NLP big model, including:

[0006] Input embedding layer: implemented using the Transformer's Embedding layer, including word embedding and position embedding sublayers; used to embed the input text and convert it into a fixed-length vector representation;

[0007] Encoder layer: includes 16 encoding sub-layers and 1 information transfer module. Each encoder sub-layer adopts the Transformer Encoder structure, and the information transfer module adopts the Cross-LayerAttention module. The information transfer module connects the 16 encoding sub-layers respectively, so that each encoding sub-layer is trained based on the same set of parameters.

[0008] Fully connected layer: used to output the prediction results of the classification task. The output dimension of the fully connected layer matches the total number of prediction results of each category task. The corresponding loss function is calculated after each iteration, and the parameter value is updated using back propagation. The loss function selects the cross entropy loss function.

[0009] Furthermore, each encoding sublayer also includes a multi-head self-attention mechanism module and a feedforward neural network. The self-attention mechanism module is used to capture the relationship between words so that the model considers other words in the sentence when processing each word; the self-attention mechanism module is connected to the feedforward neural network in sequence to further process the output of the self-attention.

[0010] Technical solution 2 adopted by the present invention:

[0011] A control method for the system described in technical solution 1, comprising the following steps:

[0012] Step 1: Data collection and preparation: Collect data related to agriculture, including soil information, meteorological data, crop growth stages, types of pests and diseases that occur, and corresponding expert solutions, and build a corpus for training the NLP model, including agricultural terms and text data in related fields;

[0013] Step 2: Data format standardization: Unify the data format to ensure that the data has a consistent structure for model training, perform data cleaning and standardization, and handle missing values ​​and outliers. When adjusting the data format, the acquired farmland-related data is used as input, including soil information, meteorological data, crop growth stage, and types of pests and diseases. The output is the corresponding expert solution, and task classification constraints are constructed so that the suggestions for problems in the farmland can be constrained to prevent uncontrollable outputs.

[0014] Step 3: Data segmentation: Split the data into training set, validation set and test set according to the preset ratio, which are used for model training, parameter adjustment and evaluation respectively;

[0015] Step 4: System training: It consists of the following specific steps:

[0016] Step 4-1: Load the training set and the test set on the system described in Technical Solution 1 and initialize them;

[0017] Step 4-2: Hyperparameter setting: set the initial values ​​of learning rate, batch size, number of iterations, and learning rate decay strategy, and load the parameters and initial weight values ​​of the pre-trained model;

[0018] Step 4-3: Pre-training system: train the model, calculate the corresponding loss after each iteration and use back propagation to update the parameter value; determine whether the predetermined end training condition is met; the end training condition is that the loss is less than the target loss, or the maximum number of training times is reached; if the end training condition is met, go to step 4-4; otherwise, go to step 4-3;

[0019] Step 4-4: Model evaluation: Use the test set to evaluate the performance of the model. The evaluation indicators include: accuracy, recall, and F1 score indicators.

[0020] The beneficial effects of the present invention are:

[0021] 1. Improve agricultural production efficiency. The system can provide farmers with precise guidance on agricultural machinery operations based on real-time data such as soil moisture, temperature, and meteorological information, combined with agricultural expertise. This helps optimize agricultural production processes and improve crop yields and quality.

[0022] 2. Use smart agricultural decision making. Through NLP big models, this enables farmers to make smarter decisions, such as asking for information about crop diseases, fertilizer recommendations, etc.

[0023] 3. Reduce the burden on farmers. The intelligence of the system can reduce the labor burden of farmers. Farmers can interact with the system without having to understand the complex technical details.

[0024] 4. Sustainable agricultural practices. Systematic guidance helps farmers adopt sustainable agricultural practices, including organic farming, water resource management, etc., to promote sustainable agricultural development.

[0025] 5. Natural language processing technology can enable farmers to interact with agricultural machinery through spoken language, thereby realizing intelligent control of agricultural machinery. In addition, natural language processing technology can also help farmers better understand the instructions for use of agricultural machinery, so as to better master the use of agricultural machinery and improve the efficiency of agricultural machinery use and the benefits of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of the structure of Example 1 of the present invention.

[0027] Figure 2 This is a flow chart of Example 2 of the present invention. DETAILED DESCRIPTION

[0028] The following examples serve to illustrate the invention.

[0029] Example 1

[0030] A real-time guidance and interaction system for agricultural machinery terminal operations based on NLP big model, including:

[0031] (1) Input embedding layer: This layer is implemented using the Transformer’s Embedding layer, which includes word embedding and position embedding sublayers. It is used to embed the input text and convert it into a fixed-length vector representation.

[0032] (2) Transformer encoder layer: It includes 16 encoding sublayers and 1 information transfer module. Each encoder sublayer adopts the Transformer encoder structure, and the information transfer module adopts the Cross-LayerAttention module. The information transfer module connects the 16 encoding sublayers respectively, so that the training of each encoding sublayer is based on the same set of parameters, which means that each layer uses the same weights. This design reduces the total number of parameters of the model while allowing the model to transfer and integrate information between different layers. The information transfer module allows information interaction between different layers, enhancing the model's ability to capture contextual information; it allows each layer to access and utilize the information learned by other layers, thereby improving the model's ability to capture long-distance dependencies. Each encoding sublayer also contains a multi-head self-attention mechanism module and a feedforward neural network, so that the model considers other words in the sentence when processing each word, thereby capturing the relationship between words; the self-attention mechanism module is connected to the feedforward neural network in sequence to further process the output of the self-attention.

[0033] Example 2

[0034] A real-time guidance and interaction method for agricultural machinery terminal operation based on NLP big model, comprising the following steps:

[0035] Step 1: Data collection and preparation: Collect data related to agriculture, including soil information, meteorological data, crop growth stages, types of pests and diseases that occur, and corresponding expert solutions, and build a corpus for training the NLP model, including agricultural terms and text data in related fields;

[0036] Step 2: Data format standardization: Unify the data format to ensure that the data has a consistent structure for model training, perform data cleaning and standardization, and handle missing values ​​and outliers. When adjusting the data format, the acquired farmland-related data is used as input, including soil information, meteorological data, crop growth stage, and types of pests and diseases. The output is the corresponding expert solution, and task classification constraints are constructed so that the suggestions for problems in the farmland can be constrained to prevent uncontrollable outputs.

[0037] Step 3: Data segmentation: Split the data into training set, validation set, and test set in a ratio of 6:2:2, which are used for model training, parameter adjustment, and evaluation respectively;

[0038] Step 4: System training: It consists of the following specific steps:

[0039] Step 4-1: Load the training set and the test set on the system described in Technical Solution 1 and initialize them;

[0040] Step 4-2: Hyperparameter setting: The learning rate is set to 1e -3 , Batch Size is set to 32, the number of iterations is set to 32, and the learning rate decay strategy is set to decay the learning rate tenfold every ten iterations; load the parameters and weights of the pre-trained model;

[0041] Step 4-3: Pre-training system: Use the optimizer Adam to train the model, calculate the corresponding loss after each iteration, and use back propagation to update the parameter value; determine whether the predetermined end training condition is met; the end training condition is that the loss is less than the target loss, or the maximum number of training times is reached; if the end training condition is met, go to step 4-4; otherwise, go to step 4-3;

[0042] Step 4-4: Model evaluation: Use the test set to evaluate the performance of the model. The evaluation indicators include: accuracy, recall, and F1 score indicators.

[0043] The operating environment configuration method of this embodiment is as follows: first download PyCharm software and Anaconda software, install a virtual environment for training in the system, use Python version 3.8, install pytorch version 2.3, and use Linux as the application system. Use pytorch to build the system described in Example 1, use the optimizer Adam to train the system, and load the prepared training set and test set.

[0044] The output dimension of this layer should match the number of categories, which is determined by the total number of expert suggestions for different problems. For each iteration, the corresponding loss is calculated and the parameter values ​​are updated using backpropagation. The loss function is the cross entropy loss function method.

[0045] The present invention integrates sensors and agricultural equipment to obtain farmland data in real time. The data includes soil information, meteorological data, crop growth stage, and the types of pests and diseases that appear. The data is uploaded to the system as input to the model. The data is used as input to the model, and the model feeds back the results to the system to obtain how to deal with the current farmland situation.

[0046] At present, the technical solution of the present invention has been piloted, and a user survey has been conducted in a small range. The survey results show that user satisfaction is high, and preparations for the transformation and promotion of the results have begun. At the same time, a survey on intellectual property risk warning has been carried out.

Claims

1. A real-time guidance and interaction system for agricultural machinery terminal operations based on NLP large model, characterized in that: include: Input embedding layer: implemented using the Transformer’s Embedding layer, including word embedding and position embedding sublayers; Used to embed the input text and convert it into a fixed-length vector representation; Encoder layer: includes 16 encoding sub-layers and 1 information transfer module. Each encoder sub-layer adopts the Transformer Encoder structure, and the information transfer module adopts the Cross-LayerAttention module. The information transfer module connects the 16 encoding sub-layers respectively, so that each encoding sub-layer is trained based on the same set of parameters. Fully connected layer: used to output the prediction results of the classification task. The output dimension of the fully connected layer matches the total number of prediction results of each category task. The corresponding loss function is calculated after each iteration, and the parameter value is updated using back propagation. The loss function selects the cross entropy loss function.

2. The real-time guidance and interaction system for agricultural machinery terminal operations based on the NLP big model according to claim 1 is characterized by: Each encoding sublayer also includes a multi-head self-attention mechanism module and a feedforward neural network. The multi-head self-attention mechanism module is used to capture the relationship between words; the self-attention mechanism module and the feedforward neural network are connected in sequence to further process the output of the self-attention.

3. A control method for the real-time guidance and interaction system for agricultural machinery terminal operations based on the NLP large model as claimed in claim 1, characterized in that: The following steps are involved: Step 1: Data collection and preparation: Collect data related to agriculture, including soil information, meteorological data, crop growth stages, types of pests and diseases that occur, and corresponding expert solutions, and build a corpus for training the NLP model, including agricultural terms and text data in related fields; Step 2: Data format standardization: Unify the data format to ensure that the data has a consistent structure for model training, perform data cleaning and standardization, and handle missing values ​​and outliers. When adjusting the data format, the acquired farmland-related data is used as input, including soil information, meteorological data, crop growth stage, and types of pests and diseases. The output is the corresponding expert solution, and task classification constraints are constructed so that the suggestions for problems in the farmland can be constrained to prevent uncontrollable outputs. Step 3: Data segmentation: Split the data into training set, validation set and test set according to the preset ratio, which are used for model training, parameter adjustment and evaluation respectively; Step 4: System training: It consists of the following specific steps: Step 4-1: Load the training set and the test set on the system described in Technical Solution 1 and initialize them; Step 4-2: Hyperparameter setting: set the initial values ​​of learning rate, batch size, number of iterations, and learning rate decay strategy, and load the parameters and initial weight values ​​of the pre-trained model; Step 4-3: Pre-training system: train the model, calculate the corresponding loss after each iteration and use back propagation to update the parameter value; determine whether the predetermined end training condition is met; the end training condition is that the loss is less than the target loss, or the maximum number of training times is reached; if the end training condition is met, go to step 4-4; otherwise, go to step 4-3; Step 4-4: Model evaluation: Use the test set to evaluate the performance of the model. The evaluation indicators include: accuracy, recall, and F1 score indicators.

Images