CN-121980973-A - Intelligent optimization method for asexual propagation of bluegrass rootstock based on big data

CN121980973ACN 121980973 ACN121980973 ACN 121980973ACN-121980973-A

Abstract

The invention relates to the technical field of propagation regulation and control, and provides an intelligent optimization method for asexual propagation of a bluegrass rhizome based on big data, which is applied to a asexual propagation production scene of the rhizome and operates in a closed-loop regulation and control system consisting of field sensing equipment, an edge computing gateway, a central intelligent decision server, a water fertilizer and agricultural machinery executing equipment. The method constructs propagation state vectors based on multisource field perception data, models a rhizome asexual propagation regulation and control process as a long-term time sequence decision problem with resource and safety constraint, introduces a prefix-guided reinforcement learning training mechanism through screening history separation strategy tracks, and realizes cooperative optimization of irrigation, fertilization and agricultural machinery operation on the premise of meeting equipment executability and agricultural safety constraint, so that stability of the rhizome asexual propagation of bluegrass and resource utilization efficiency are improved.

Inventors

Pang Xiaopan
LI JIENA
WEI XIAOXING
WANG QIAN
WANG XIAOZHENG

Assignees

兰州大学

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (8)

1. An intelligent optimization method for the rhizome asexual propagation of bluegrass based on big data is characterized by comprising the following steps: S1, collecting original breeding data; s2, processing the original breeding data to form a breeding state sequence; Step S3, acquiring a rootstock asexual propagation yield index, and mapping the rootstock asexual propagation yield index into a reward signal; Step S4, defining propagation regulation actions and establishing action constraint sets; step 5, constructing a strategy-leaving track library, wherein each strategy-leaving track in the strategy-leaving track library consists of a propagation state sequence and a propagation regulation action, and is associated with a track period return value obtained by recording a reward signal calculation; step S6, constructing a reinforcement learning model comprising a strategy network and a value network, obtaining a return and advantage estimation result, and introducing a prefix guiding mechanism and a dynamic training proportion of a prefix-free task to update parameters of the strategy network to obtain a trained strategy network; And S7, calling the trained strategy network to output the current reproduction control action, and after the feasibility verification of the action constraint set, issuing the action constraint set to an electromagnetic valve, a variable frequency pump, a valve group, a fertilization pump, a fertilizer mixer and an operation scheduling terminal, and writing the execution result and the output index back to the strategy-leaving track library.
2. The intelligent optimization method for the asexual propagation of the roots and stems of the bluegrass based on big data, which is characterized in that the reward signal comprises a yield improvement item and a resource cost penalty item, the resource cost penalty item comprises a water consumption penalty, a fertilization amount penalty and a super-threshold safety penalty item, and the super-threshold safety penalty item is triggered when any situation that the water content of soil exceeds an upper limit threshold or the conductivity exceeds a salt damage threshold or nitrate nitrogen exceeds a leaching risk threshold occurs.
3. The intelligent optimization method for the asexual propagation of the bluegrass rootstock based on big data is characterized by comprising the steps of constructing an off-strategy track library, wherein each off-strategy track in the off-strategy track library consists of a propagation state sequence and a propagation regulation action corresponding to the asexual propagation process of the bluegrass rootstock, and is associated with a track periodic return value obtained by calculating a record rewarding signal, screening each track in the off-strategy track library according to a preset propagation effect index to obtain an effective track set, defining propagation tasks, selecting a corresponding correct track for each propagation task in the effective track set, extracting a propagation action sequence in the correct track, encoding the propagation action sequence into an operation mark sequence, constructing a candidate prefix length set based on the operation mark sequence, intercepting the operation mark sequence in the candidate prefix length set according to different prefix lengths to obtain track prefix segments, splicing the track prefix segments with task representations of the corresponding propagation tasks to obtain a task representation with prefixes, and generating a prefix set by setting a plurality of different prefix lengths.
4. The intelligent optimization method for the asexual propagation of the bluegrass rhizome based on big data, which is characterized by comprising the steps of establishing a baseline strategy model according to the following rule, taking a periodic return predicted value of the baseline strategy model under the condition of a given track prefix section as a prefix selection condition, adopting a segmentation candidate prefix length selection mechanism, respectively calculating the periodic return predicted value of the baseline strategy model under the condition of a corresponding prefix section in a candidate prefix length set, selecting a candidate prefix length section with the periodic return predicted value being in jump as an effective prefix length section for generating a prefix problem set, and determining the prefix length.
5. The intelligent optimization method for the asexual propagation of the bluegrass rootstock based on big data as claimed in claim 3, wherein the propagation effect index comprises one or more of rootstock expansion rate, seedling density per unit area, tillering node formation rate, seedling uniformity and fertilizer utilization gain.
6. The intelligent optimization method for the asexual propagation of the bluegrass rootstock based on big data, which is characterized by comprising the steps of introducing a propagation consistency threshold value to screen each separation strategy track in a separation strategy track library on the basis of meeting propagation effect indexes, wherein the propagation consistency threshold value is obtained by combined calculation of seedling density dispersion indexes and seedling uniformity dispersion indexes of different grid units in the same land.
7. The intelligent optimization method for the asexual propagation of the roots and stems of the bluegrass based on big data, which is characterized by comprising the following steps of: step S61, constructing a reinforcement learning model comprising a strategy network and a value network, wherein the strategy network is used for generating subsequent asexual propagation regulation and control actions of the roots and stems of the bluegrass under the given propagation task condition, and the value network is used for estimating propagation period returns under the current strategy condition; step S62, acquiring a prefix-free task set, and simultaneously taking the prefix problem set and the prefix-free task set as input execution strategy sampling in a training stage based on the reinforcement learning model of the step S61 to acquire a corresponding periodic return and advantage estimation result; when the strategy updating is carried out on the training samples corresponding to the prefix problem set, a zero weight mask is set for the loss items corresponding to the track prefix section through a prefix guiding mechanism, so that the gradient is not returned through the track prefix section, strategy loss, value loss and entropy regular items are calculated only for the subsequent decision section generated after the track prefix section, and the strategy network and value network parameters are jointly updated by combining the updating result of the prefix-free task samples of the prefix task set; Step S63, in the process of updating the combined strategy of the step S62, the sampling proportion of the prefix problem set and the prefix-free task set in training is dynamically adjusted based on the progress state of the training stage, the sampling proportion of the prefix problem set is increased in the initial stage of training to improve the acquisition efficiency of a high-return sample, the sampling proportion of the prefix-free task set is gradually increased in the later stage of training to strengthen the autonomous decision making capability of the learning model under the prefix-free condition, and the trained strategy network is output after the preset training round condition is met.
8. The intelligent optimization method for the asexual propagation of the bluegrass rhizome based on big data, which is characterized in that the off-strategy track library is stored by adopting a seasonal and plot index structure.

Description

Intelligent optimization method for asexual propagation of bluegrass rootstock based on big data Technical Field The invention relates to the technical field of propagation regulation and control, in particular to an intelligent optimization method for asexual propagation of bluegrass rootstock based on big data. Background The bluegrass is used as a typical cold season type turf grass and pasture variety, and is generally produced in a large scale by adopting a rhizome asexual propagation mode in a turf production base, a pasture propagation base and an agricultural scientific research test field, wherein the propagation mode has high sensitivity to soil moisture state, nutrient supply structure, illumination condition, temperature environment, operation time sequences such as trimming, earthing and compacting, and the like, and the propagation process has the characteristics of multi-source environment factor coupling, multi-decision variable linkage, time sequence nonlinear evolution and obvious space difference. In actual production, yield indexes such as rhizome extension length, tiller node formation quantity, seedling density, seedling alignment, overground biomass and the like dynamically change along with propagation stages and environmental conditions, and obvious stage dependency and hysteresis effect exist between water and fertilizer input strength and operation parameters, so that a single model or static strategy is difficult to uniformly describe a complete propagation period. Along with the application of field sensing equipment and an agricultural information system, multisource sensing data is gradually accumulated, but the existing method focuses on the analysis and prediction of single environmental indexes or short-term states, lacks system modeling of mapping relation between propagation regulation and control behaviors and long-term output return, particularly lacks a structural utilization mechanism of high-quality decision sequences in a historical propagation process, and is difficult to realize stable optimization and cross-season strategy iteration of the whole process of rhizome asexual propagation under a multi-constraint condition. Disclosure of Invention According to the intelligent optimization method for the asexual propagation of the bluegrass rootstock, aiming at the characteristics that the environmental state is highly coupled, the regulation and control actions have obvious time sequence dependence and the propagation output is highly sensitive to a key operation stage in the asexual propagation process of the bluegrass rootstock, the intelligent optimization method for the asexual propagation of the bluegrass rootstock is provided, the track prefix which covers the stages of irrigation strategy switching, fertilization ratio adjustment, pruning operation window determination, height setting and the like is extracted from a closed loop system with cooperative multi-source perception, edge calculation and central intelligent decision in the field, unified modeling is carried out on multidimensional information such as soil moisture, nutrient supply, illumination conditions, meteorological driving and operation history and the like, a propagation state vector capable of continuously representing the evolution characteristics of the propagation process is constructed, the asexual propagation regulation and control problem of the bluegrass rootstock is formed into a long-term time sequence decision problem with resource constraint and safety constraint, on the basis of introducing a separation strategy track prefix modeling and guiding mechanism which faces propagation tasks, the stable propagation effect and good space consistency in the history production season are carried out, the track prefix of the asexual propagation effect is extracted from the closed loop system, the track prefix which covers the stages of irrigation strategy switching, the fertilization ratio adjustment, the pruning operation window determination and the height setting and the like is carried out, the condition information is embedded into the training process, the condition information, the asexual propagation strategy is carried out by the pre-training mode, the strategy is maintained in the advanced strategy and the advanced operation strategy and the asexual operation strategy is superior to the asexual operation strategy and the operation strategy and has high iteration performance and has high performance and high performance. The invention provides an intelligent optimization method for the asexual propagation of the roots and stems of bluegrass based on big data, which is executed in a closed-loop regulation and control system consisting of field sensing equipment, an edge computing gateway, a central intelligent decision server, water fertilizer and agricultural machinery execution equipment, and comprises the following steps: Step S1, multisource breeding collection and calibration, na