CN-121981451-A - Fig water and fertilizer digital management method based on low-code platform

CN121981451ACN 121981451 ACN121981451 ACN 121981451ACN-121981451-A

Abstract

The invention relates to the technical field of intelligent agriculture, in particular to a fig water and fertilizer digital management method based on a low-code platform. The method comprises the steps of constructing a water and fertilizer decision model integrating time sequence prediction and an attention mechanism through a low-code platform, processing an environment data sequence through an improved transducer architecture, capturing periodic characteristics by combining a learnable time perception code, establishing a segmentation attention weight distribution strategy based on a growth stage, dynamically adjusting attention degrees of different environment characteristics, configuring a model updating system comprising an online learning mechanism, and realizing continuous optimization of model parameters through an elastic weight consolidation algorithm. The method realizes the accurate prediction and self-adaptive regulation of the fig water and fertilizer requirements, and effectively improves the water and fertilizer management precision and the system usability.

Inventors

ZHANG QIANG
CHEN XIANGYU
LI YU
LEI YAN
CAO BOBO
LIU HAIYAN

Assignees

石河子大学

Dates

Publication Date: 20260505
Application Date: 20260111

Claims (10)

1. The fig water and fertilizer digital management method based on the low-code platform is characterized by comprising the following steps of: S1, constructing a water and fertilizer decision model through a graphical configuration interface of a low-code platform, wherein the water and fertilizer decision model comprises a time sequence prediction module and an attention mechanism module; S2, collecting historical environmental data and crop growth data of the fig planting area, wherein the historical environmental data comprise soil humidity, soil temperature, air humidity and illumination intensity, and the crop growth data comprise leaf form data and fruit development data; s3, inputting the collected data into a time sequence prediction module, and predicting the water and fertilizer requirement change trend of the fig tree within 5-30 days in the future; S4, dynamically adjusting weight distribution of different input features in the prediction process through an attention mechanism module, and focusing on features related to key growth stages of figs; s5, generating a water and fertilizer control instruction based on a prediction result with an attention mechanism; S6, sending a water and fertilizer control instruction to water and fertilizer execution equipment to control irrigation and fertilizer application operation.
2. The fig water and fertilizer digitization management method based on a low-code platform according to claim 1, wherein in step S1, the time-series prediction module employs a modified transform architecture, in which an encoder section contains 6-12 encoding layers, each layer is equipped with a self-attention mechanism for processing historical environmental data sequences, a decoder section contains 4-8 decoding layers, each layer is equipped with a cross-attention mechanism for generating future water and fertilizer demand prediction sequences, and a user can configure the number of encoding layers and decoding layers, and the number of attention heads through the low-code platform.
3. The fig water and fertilizer digital management method based on the low-code platform according to claim 2, wherein the position code in the improved transducer architecture adopts a leachable time-aware code, and the code comprises a first part based on the absolute position of data in a sequence and a second part based on a seasonal period and a fig growth period, wherein the seasonal period is set to 365 days, the fig growth period is set to 120-180 days according to the variety characteristics, and the user can adjust the period parameters through the low-code platform.
4. The fig water and fertilizer digital management method based on the low-code platform according to claim 1, wherein in step S4, the attention mechanism module comprises a segmented attention weight distribution strategy based on fig growth stages, the strategy divides the fig growth period into 4 main stages, namely a germination stage, a flowering stage, a fruit expansion stage and a maturation stage, different attention weight preference is set in each stage, so that the model focuses on soil temperature and air temperature in the germination stage, focuses on soil humidity and illumination intensity in the flowering stage, focuses on soil humidity and air temperature in the fruit expansion stage, and focuses on illumination intensity and soil humidity in the maturation stage.
5. The fig water and fertilizer digital management method based on the low-code platform according to claim 4, wherein the segmented attention weight distribution strategy is realized in the form of configurable attention templates through the low-code platform, each template comprises attention weight preset configuration for a specific growth stage, the weight value ranges from 0.1 to 1.0, and a user can copy and modify the templates according to planting experience and adjust the attention weight of each feature through a graphical slider.
6. The fig water and fertilizer digital management method based on the low-code platform according to claim 1, wherein the method further comprises a model online learning step S7, wherein the system collects new environment data, crop growth data and actual water and fertilizer application effect data every 7-15 days, the parameters of a time sequence prediction model are updated through an incremental learning algorithm, the incremental learning adopts an elastic weight consolidation algorithm, model parameters important to a historical prediction task are identified through an attention mechanism, protection constraint is applied to the parameters, and a constraint intensity coefficient is set to be 0.5-2.0.
7. The fig water and fertilizer digital management method based on a low-code platform according to claim 6, wherein in step S7, the system analyzes the influence degree of new data on the model prediction accuracy through an attention mechanism, when the attention weight distribution of the new data and the difference between the historical modes exceed a set threshold by 15% -30%, a model parameter updating process is automatically triggered, and a user can adjust the difference threshold and the model updating frequency through the low-code platform.
8. The fig water and fertilizer digital management method based on the low-code platform according to claim 1 is characterized in that the time sequence prediction module adopts a multi-task learning framework and simultaneously executes 3 prediction tasks, namely short-term water and fertilizer demand prediction in 7 days in the future, long-term growth trend prediction in 30-60 days in the future and abnormal situation risk prediction, wherein the attention mechanism is shared among different prediction tasks, but generates different attention weight distribution for each task, and a user can adjust the loss function weight of each task through the low-code platform, wherein the weight range is 0.1-1.0.
9. The fig water and fertilizer digital management method based on the low-code platform is characterized in that the abnormal condition risk prediction task is used for specifically detecting a water and fertilizer stress early signal, the task uses an attention mechanism to highlight an abnormal environment data mode, when the soil humidity is detected to be lower than a set threshold value for 3-5 days continuously and the attention weight is higher than 0.7, the system generates an early warning signal, and a user can adjust the early warning threshold value and the detection sensitivity through the low-code platform.
10. The fig water and fertilizer digital management method based on the low-code platform according to claim 1 is characterized in that the low-code platform provides an attention visualization component which displays attention weight distribution of different input features in a prediction process in a thermodynamic diagram form, the color shade of the thermodynamic diagram represents the weight value, the weight value ranges from 0.1 to 1.0, a user can manually adjust the attention weight through an interactive interface, the adjustment range is limited within +/-20% of an original weight, and the manual adjustment is used as a constraint condition to be integrated into the next training process of a model.

Description

Fig water and fertilizer digital management method based on low-code platform Technical Field The invention relates to the technical field of intelligent agriculture, in particular to a fig water and fertilizer digital management method based on a low-code platform. Background Fig planting is taken as a special agricultural industry, and the water and fertilizer management precision of the fig planting directly influences the quality and yield of fruits. Along with the development of intelligent agricultural technology, the traditional water and fertilizer management mode depending on experience is difficult to meet the accurate and intelligent production requirements. The establishment of a scientific water and fertilizer decision system has important significance for improving economic benefit and sustainable development of fig planting. The prior fig water and fertilizer digital management mainly has the defects that firstly, the prior decision model is mostly a static rule system, a fixed threshold value is set depending on expert experience, and the prior decision model cannot adapt to the dynamic change of crop requirements in different growth stages, so that the water and fertilizer application time and the water and fertilizer application dosage are not accurate enough. Secondly, the traditional model lacks depth mining of inherent association of multi-source data, and complex nonlinear relations between environmental data and crop physiological states cannot be effectively analyzed, so that prediction accuracy is affected. And moreover, the existing system generally has the problems of high specialization degree and complex operation, and agricultural technicians are difficult to adjust and optimize model parameters according to actual planting conditions. In addition, most systems adopt a fixed algorithm architecture, the evolution cannot be continuously learned through the use process, and the performance of the model gradually decreases along with the change of the planting environment. Finally, existing solutions typically implement data collection, analysis, and decision-making as independent modules, lacking complete digital workflow support, resulting in slow system response and limited management efficiency. Therefore, the fig water and fertilizer digital management method based on the low-code platform is provided for solving the core problem of how to construct a water and fertilizer decision system capable of adapting to the change of fig growth requirements, so that the use threshold of the system is reduced while the prediction precision of a model is ensured, and agricultural technicians can conveniently participate in the model optimization process to realize continuous improvement of water and fertilizer management. Disclosure of Invention In order to overcome the defects in the prior art, the embodiment of the invention provides a fig water and fertilizer digital management method based on a low-code platform, which aims to solve the problems in the background art. In order to achieve the purpose, the invention provides the technical scheme that the fig water and fertilizer digital management method based on the low-code platform comprises the following steps of: S1, constructing a water and fertilizer decision model through a graphical configuration interface of a low-code platform, wherein the water and fertilizer decision model comprises a time sequence prediction module and an attention mechanism module; S2, collecting historical environmental data and crop growth data of the fig planting area, wherein the historical environmental data comprise soil humidity, soil temperature, air humidity and illumination intensity, and the crop growth data comprise leaf form data and fruit development data; s3, inputting the collected data into a time sequence prediction module, and predicting the water and fertilizer requirement change trend of the fig tree within 5-30 days in the future; S4, dynamically adjusting weight distribution of different input features in the prediction process through an attention mechanism module, and focusing on features related to key growth stages of figs; s5, generating a water and fertilizer control instruction based on a prediction result with an attention mechanism; S6, sending a water and fertilizer control instruction to water and fertilizer execution equipment to control irrigation and fertilizer application operation. Preferably, in step S1, the time series prediction module employs a modified transform architecture, wherein the encoder section comprises 6-12 encoding layers, each layer is equipped with a self-attention mechanism for processing historical environmental data sequences, the decoder section comprises 4-8 decoding layers, each layer is equipped with a cross-attention mechanism for generating future water and fertilizer demand prediction sequences, the number of encoding layers and decoding layers, and the number of attention heads are confi