A method for maintaining activity of pepper fruit VIGS based on data analysis

Abstract

This invention discloses a data analysis-based method for maintaining VIGS activity in chili pepper fruits, belonging to the technical field of VIGS activity maintenance in chili pepper fruits. The method includes the following steps: Multimodal data acquisition: After VIGS inoculation of chili pepper fruits, a multimodal sensor array is deployed to continuously collect environmental factor data of the microenvironment where the fruit is located at a preset first frequency. This invention, by deploying a multimodal sensor array, achieves real-time and continuous acquisition of microenvironmental factors and physiological state data of chili pepper fruits. Combined with near-infrared spectroscopy and a self-calibration mechanism, it significantly improves the estimation accuracy of the target gene silencing efficiency index and the robustness of data acquisition. Based on a nonlinear mixed-effects model and Bayesian parameter inference method, this invention constructs an individualized dynamic decay model of VIGS silencing activity, which can update the silencing efficiency decay curve of individual fruits in real time, accurately predict the remaining maintenance time, and improve the targeting and effectiveness of regulation.

Description

Technical Field

[0001] This invention relates to the field of VIGS activity maintenance technology in chili pepper fruits, and more specifically, to a method for maintaining VIGS activity in chili pepper fruits based on data analysis. Background Technology

[0002] Virus-induced gene silencing (VIGS) technology has been widely used in chili pepper fruit gene function research due to its advantages such as ease of operation, short cycle, and no need for genetic transformation. By inoculating the fruit with a viral vector during critical fruit development stages, the target gene can be transiently silenced, thereby exploring its biological function in fruit ripening, stress response, and other processes. However, the VIGS silencing efficiency is not constant and is affected by multiple factors such as fruit physiological state, microenvironmental factors (e.g., temperature, humidity, light), and individual differences. The silencing activity often gradually declines after inoculation, severely limiting the effective utilization of the experimental window and the reliability of the data.

[0003] In existing technologies, maintaining VIGS silencing efficiency mainly relies on manual observation and empirical regulation. Researchers typically set fixed environmental parameters according to predetermined experimental protocols or passively intervene when a decline in silencing effect is observed. This regulatory approach has significant limitations: First, due to the lack of real-time perception of individual fruit physiological states and microenvironmental changes, regulatory decisions often lag behind actual needs, making it difficult to intervene effectively in the early stages of silencing efficiency decline. Second, traditional methods cannot quantify and assess the immediate and cumulative effects of different regulatory measures on silencing efficiency, resulting in significant uncertainty and randomness in the regulatory effects. Third, manual regulation struggles to achieve multi-objective synergistic optimization, such as controlling energy consumption while maintaining silencing efficiency and avoiding physiological stress on the fruit caused by over-regulation, leading to low resource utilization efficiency and even affecting normal fruit development and the reliability of experimental results. Furthermore, existing technologies lack systematic modeling and prediction capabilities for the dynamic changes in VIGS silencing activity, failing to provide personalized regulatory strategies for fruits of different varieties, batches, and environmental conditions, severely restricting the widespread application of VIGS technology in pepper fruit gene function research and the reproducibility of experimental data.

[0004] Based on this, the present invention designs a data analysis-based method for maintaining VIGS activity in chili pepper fruits to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a data analysis-based method for maintaining VIGS activity in chili pepper fruits, in order to solve the problems mentioned in the background art.

[0006] A data analysis-based method for maintaining VIGS activity in chili pepper fruits includes the following steps: S1. Multimodal Data Acquisition: After VIGS inoculation of pepper fruits, a multimodal sensor array is deployed to continuously acquire environmental factor data of the microenvironment where the fruit is located at a preset first frequency, and to continuously acquire physiological state data of the fruit at a preset second frequency. The environmental factor data includes at least temperature, humidity, light intensity, and... The concentration, the physiological state data includes at least the water potential of fruit tissue, cell membrane permeability, and the relative expression level of target gene mRNA determined by minimally invasive sampling combined with quantitative PCR technology; S2. State Quantification and Decay Model Construction: Based on the real-time acquisition of the relative expression level of target gene mRNA, the VIGS silencing efficiency index at the current moment is calculated. With time as the independent variable and the silencing efficiency index as the dependent variable, a dynamic decay model of the silencing activity of individual fruits is constructed by fitting and predicting the remaining maintenance time required for the silencing efficiency index to decay to a preset threshold. S3. Feedforward-feedback composite control: The environmental factor data, fruit tissue water potential, cell membrane permeability, and remaining maintenance time are used as input variables and input into a pre-trained state-action value network. The state-action value network is constructed based on a deep reinforcement learning model and aims to maximize the duration for which the silencing efficiency index is higher than the preset threshold and minimize the energy consumption of environmental regulation. The state-action value network outputs the optimal regulation action at the current moment. The regulation action includes adjusting the environmental factor setpoint or starting / stopping the application of exogenous hormones. S4. Personalized control execution: Based on the optimal control action, a control command is generated and sent to the environmental control execution mechanism to dynamically adjust the fruit microenvironment. At the same time, steps S1 to S4 are repeated at a preset third frequency to form a closed-loop control until the remaining maintenance time is zero.

[0007] Preferably, the specific process of constructing the dynamic decay model of the silencing activity of individual fruits in step S2 is as follows: Based on a nonlinear mixed-effects model, the change in VIGS silencing efficiency over time is decomposed into a fixed-effects component and a random-effects component. The fixed-effects component describes the average decay of silencing efficiency over time at the population level, while the random-effects component describes the degree to which individual fruits deviate from the population average due to differences in inoculation and physiological state. Using a Bayesian parameter inference method, the random-effects parameters of individual fruits are updated online using real-time collected data on the relative expression levels of target gene mRNAs. This generates a personalized dynamic decay model of silencing activity in real time, and the remaining maintenance time is predicted based on this model.

[0008] Preferably, the training process of the state-action value network in step S3 includes: A computational simulation environment was constructed to simulate the VIGS silencing process under different environmental disturbances and fruit physiological states. In this simulation environment, a dynamic decay model of silencing activity driven by historical experimental data was used as the core. A state space, action space, and reward function were defined. The reward function consisted of a silencing efficiency maintenance reward term and an energy consumption penalty term. Through reinforcement learning algorithms, the agent was driven to conduct massive trial-and-error learning in the simulation environment to maximize the cumulative reward, thereby obtaining the state-action value network that selects the optimal regulatory action under different states. After training, the state-action value network was deployed to the actual regulatory system for online decision-making.

[0009] Preferably, in step S1, a near-infrared spectral sensor is used instead of the minimally invasive sampling combined with quantitative PCR technology to collect the characteristic spectral data of the fruit in real time in a non-contact manner, and the characteristic spectral data is converted into an equivalent target gene silencing efficiency index based on a pre-established spectral-silencing efficiency correlation model.

[0010] Preferably, the optimal control action in step S3 is generated based on a multi-objective optimization algorithm. The multi-objective optimization algorithm takes maintaining the silencing efficiency index within a preset qualified range, maintaining the fruit tissue water potential within a preset healthy range, and minimizing the energy consumption of environmental control as the collaborative optimization objectives. When the silencing efficiency index is higher than the upper limit of the preset qualified range, the control action prioritizes reducing the intensity of environmental control rather than starting hormone spraying, so as to avoid causing excessive stress to the fruit.

[0011] Preferably, the state-action value network in step S3 includes two stages in its training process: offline pre-training and online adaptive updating. In the offline pre-training stage, an initial network model is trained using experimental data containing multiple chili varieties and historical environmental conditions. In the online adaptive updating stage, when the system is deployed on a new fruit population, the silencing efficiency data and environmental response data of the current batch are collected in real time, and a meta-learning algorithm is used to quickly fine-tune some parameters of the initial network model to adapt it to the dynamic characteristics of the current variety and environment. The updated model is used for subsequent regulatory decisions.

[0012] Preferably, step S1 further includes a self-calibration step for the spectral sensor: a reference standard with stable spectral reflectance characteristics is fixedly set next to the fruit sample. Before each collection of fruit spectral data, the spectral signal of the reference standard is first collected, the deviation between the reference standard and the standard reference spectrum is calculated, and the subsequent collected fruit characteristic spectral data is corrected in real time according to the deviation to eliminate systematic errors introduced by sensor drift and environmental fluctuations. At the same time, the spectral-silence efficiency correlation model is constructed using an ensemble learning algorithm, which integrates the outputs of multiple sub-models of different bands. When a certain band signal is disturbed, its weight is automatically reduced to ensure the robustness of the silence efficiency index estimation.

[0013] Preferably, the multi-objective optimization algorithm is a model predictive control algorithm. This algorithm not only optimizes based on the current state, but also makes rolling predictions of the fruit state within a preset time window based on the dynamic decay model of silencing activity. Within the prediction time window, the algorithm comprehensively evaluates the cumulative impact of the current control action on future silencing efficiency, fruit health, and environmental energy consumption, selects the control action sequence that maximizes the comprehensive benefits within the entire prediction window, and executes the first action of the sequence as the optimal control action at the current moment, thereby achieving proactive response and preventive control to environmental disturbances.

[0014] Preferably, the online adaptive update phase includes an update protection mechanism: before each parameter fine-tuning of the initial network model, the distribution similarity between the current batch of real-time data and the offline training dataset is first calculated; when the distribution similarity is lower than a preset threshold, the online update of the model parameters is paused, and a rule-based control strategy is adopted as a backup scheme to output control actions. A rapid data acquisition process is initiated based on the current batch of data. After the amount of acquired data meets the requirements, the distribution similarity is re-evaluated to determine whether to resume model updates. At the same time, during the model update process, an elastic weight solidification algorithm is introduced to impose a high update penalty on network parameters that affect important features of historical data to prevent catastrophic forgetting.

[0015] Compared with the prior art, the advantages of this invention are: 1. This invention achieves real-time and continuous acquisition of data on microenvironmental factors and physiological states of pepper fruits by deploying a multimodal sensor array. Combined with near-infrared spectroscopy and a self-calibration mechanism, it significantly improves the estimation accuracy of the target gene silencing efficiency index and the robustness of data acquisition.

[0016] 2. Based on the nonlinear mixed-effects model and Bayesian parameter inference method, this invention constructs an individualized VIGS silencing activity dynamic decay model, which can update the silencing efficiency decay curve of individual fruits in real time, accurately predict the remaining maintenance time, and improve the targeting and effectiveness of regulation.

[0017] 3. This invention introduces deep reinforcement learning to construct a state-action value network, combined with a feedforward-feedback composite control strategy, with the optimization objectives of maximizing silencing efficiency maintenance time and minimizing energy consumption, to achieve intelligent decision-making and dynamic regulation of the VIGS activity maintenance process, significantly improving regulation efficiency and resource utilization.

[0018] 4. This invention uses multi-objective optimization algorithms (such as model predictive control) and priority regulation strategies to maintain silencing efficiency while taking into account fruit health and environmental energy consumption, avoiding physiological stress caused by over-regulation, and improving fruit survival rate and experimental success rate.

[0019] 5. This invention adopts a model training mechanism that combines offline pre-training with online adaptive updating, and combines meta-learning and elastic weight solidification algorithms to enable the system to have good generalization ability and anti-interference ability, adapt to changes in different varieties, environments and batches, and has strong practicality.

[0020] 6. This invention sets up an update protection mechanism, introduces distribution similarity evaluation and rule-based backup strategy, to ensure that the system can still operate stably when the data distribution changes drastically, prevents catastrophic model forgetting and misregulation, and significantly improves the security and robustness of the system.

[0021] 7. This invention achieves continuous optimization and dynamic adjustment of the control action through a closed-loop feedback control mechanism until the VIGS activity maintenance period ends, forming a complete intelligent closed-loop control process, which improves the repeatability and data reliability of VIGS experiments. Attached Figure Description

[0022] Figure 1 The present invention provides a process flow for a data analysis-based method for maintaining VIGS activity in chili pepper fruits. Figure 1 ; Figure 2 The present invention provides a process flow for a data analysis-based method for maintaining VIGS activity in chili pepper fruits. Figure 2 . Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figures 1-2 A data analysis-based method for maintaining VIGS activity in chili pepper fruits includes the following steps: S1. Multimodal Data Acquisition: After VIGS inoculation of pepper fruits, a multimodal sensor array is deployed to continuously acquire environmental factor data of the fruit's microenvironment at a preset first frequency, and physiological state data of the fruit at a preset second frequency. The environmental factor data includes at least temperature, humidity, light intensity, and... Concentration and physiological state data include at least the water potential of fruit tissue, cell membrane permeability, and the relative expression level of target gene mRNA determined by minimally invasive sampling combined with quantitative PCR technology. The multimodal sensor array includes: Environmental factor sensors include temperature and humidity sensors (such as SHT35, accuracy ±0.2℃, ±1.5%RH), light sensors (such as LI-190R, measurement range 400-700nm), and CO2 sensors (such as GSSS8, measurement range 0-5000ppm). The data acquisition frequency is preset to once every 5 minutes (first frequency). Physiological state sensors: A miniature pressure chamber (such as the PMS Model 600) is used to measure the water potential of fruit tissue; a conductivity meter (such as the Mettler Toledo FiveEasy) is used to measure cell membrane permeability; the relative expression level of target gene mRNA is measured using minimally invasive sampling combined with qRT-PCR technology, with the sampling frequency preset to once every 12 hours (second frequency). To improve temporal resolution, a near-infrared spectral sensor (such as the Viavi MicroNIR 1700) can also be used to replace qRT-PCR, establishing a spectral-silencing efficiency model to achieve non-contact, real-time estimation.

[0025] After data collection, preprocessing is required, including outlier removal (based on the 3σ principle), missing value imputation (using linear interpolation or forward imputation), and data normalization (Z-score standardization) to facilitate subsequent model input. S2. State Quantification and Decay Model Construction: Based on the real-time acquisition of the relative expression level of target gene mRNA, the VIGS silencing efficiency index at the current moment is calculated. With time as the independent variable and the silencing efficiency index as the dependent variable, a dynamic decay model of the silencing activity of individual fruits is constructed and the remaining maintenance time required for the silencing efficiency index to decay to the preset threshold is predicted. To quantify the decay pattern of VIGS silencing activity in individual fruits, the following nonlinear mixed-effects model was constructed: Fixed effects section: using the logistic decay function Describes the average level of a group; Random effects section: Introducing individual random effects parameters Describes an individual's maximum silencing efficiency Or a deviation on the decay rate k; Bayesian parameter inference: The Hamiltonian Monte Carlo (HMC) method is used, combined with real-time collected individual expression data, to update individual random effects online, generate personalized decay curves in real time, and predict the remaining maintenance time required for the silencing efficiency to drop to a preset threshold (e.g., 50%). ; S3. Feedforward-feedback composite control: Environmental factor data, fruit tissue water potential, cell membrane permeability, and remaining maintenance time are used as input variables and input into a pre-trained state-action value network. The state-action value network is constructed based on a deep reinforcement learning model and aims to maximize the duration for which the silencing efficiency index is higher than a preset threshold and minimize the energy consumption of environmental regulation. The state-action value network outputs the optimal regulation action at the current moment. The regulation action includes adjusting the environmental factor setpoint or starting / stopping the spraying of exogenous hormones. S4. Personalized control execution: Based on the optimal control action, control instructions are generated and sent to the environmental control execution mechanism to dynamically adjust the fruit microenvironment. At the same time, steps S1 to S4 are repeated at a preset third frequency to form a closed-loop control until the remaining maintenance time is zero.

[0026] The specific methods for dynamically adjusting the fruit microenvironment are as follows: Based on the optimal control action, control instructions are generated and sent to the environmental control execution mechanism to achieve dynamic adjustment of environmental factors such as temperature, humidity, light intensity and CO2 concentration in the fruit microenvironment, or to control the start and stop of exogenous hormone spraying.

[0027] The specific process of constructing the dynamic decay model of the silencing activity of individual fruits in step S2 is as follows: Based on a nonlinear mixed-effects model, the change in VIGS silencing efficiency over time is decomposed into a fixed-effects component and a random-effects component. The fixed-effects component describes the average decay of silencing efficiency over time at the population level, while the random-effects component describes the degree to which individual fruits deviate from the population average due to differences in inoculation and physiological state. Using Bayesian parameter inference methods, the random-effects parameters of individual fruits are updated online using real-time data on the relative expression levels of target gene mRNAs. This generates a personalized dynamic decay model of silencing activity in real time, and the remaining maintenance time is predicted based on this model.

[0028] The state-action value network in step S3, its training process includes: A computational simulation environment was constructed to simulate the VIGS silencing process under different environmental disturbances and fruit physiological states. In the simulation environment, a dynamic decay model of silencing activity driven by historical experimental data was used as the core. The state space, action space, and reward function were defined. The reward function consists of a silencing efficiency maintenance reward term and an energy consumption penalty term. Through reinforcement learning algorithm, the agent was driven to conduct massive trial and error learning in the simulation environment to maximize the cumulative reward, thereby obtaining a state-action value network for selecting the optimal regulatory action under different states. After training, the state-action value network was deployed to the actual regulatory system for online decision-making.

[0029] In step S1, a near-infrared spectral sensor is used to replace minimally invasive sampling combined with quantitative PCR technology. Characteristic spectral data of the fruit are collected in real time in a non-contact manner. Based on a pre-established spectral-silencing efficiency correlation model, the characteristic spectral data are converted into an equivalent target gene silencing efficiency index.

[0030] The optimal control action in step S3 is generated based on a multi-objective optimization algorithm. The multi-objective optimization algorithm takes maintaining the silencing efficiency index within a preset qualified range, maintaining the fruit tissue water potential within a preset healthy range, and minimizing the energy consumption of environmental control as the collaborative optimization objectives. When the silencing efficiency index is higher than the upper limit of the preset qualified range, the control action prioritizes reducing the intensity of environmental control rather than starting hormone spraying, so as to avoid excessive stress on the fruit.

[0031] The state-function value network in step S3 has two training stages: offline pre-training and online adaptive updating. In the offline pre-training stage, an initial network model is trained using experimental data containing multiple chili varieties and historical environmental conditions. In the online adaptive updating stage, when the system is deployed to a new fruit population, the silencing efficiency data and environmental response data of the current batch are collected in real time. A meta-learning algorithm is used to quickly fine-tune some parameters of the initial network model to adapt it to the dynamic characteristics of the current variety and environment. The updated model is used for subsequent regulatory decisions.

[0032] Step S1 also includes a self-calibration step for the spectral sensor: a reference standard with stable spectral reflectance characteristics is fixed next to the fruit sample. Before each collection of fruit spectral data, the spectral signal of the reference standard is first collected, the deviation between the reference standard and the standard reference spectrum is calculated, and the subsequent fruit characteristic spectral data is corrected in real time based on the deviation to eliminate systematic errors introduced by sensor drift and environmental fluctuations. At the same time, the spectral-silence efficiency correlation model is constructed using an ensemble learning algorithm, which integrates the outputs of multiple sub-models in different bands. When a certain band signal is disturbed, its weight is automatically reduced to ensure the robustness of the silence efficiency index estimation.

[0033] The reference standard with stable spectral reflectance characteristics consists of a diffuse reflectance white plate made of polytetrafluoroethylene (PTFE) material pressed at high temperature, encapsulated in a metal aluminum box with a protective threaded cap. This standard is installed next to a pepper fruit sample, with its reflective surface facing the spectral sensor. This allows the sensor to acquire the spectral signal from the standard for real-time calibration before each collection of fruit spectral data.

[0034] The multi-objective optimization algorithm is a model predictive control algorithm. This algorithm not only optimizes based on the current state, but also makes rolling predictions of the fruit state within a preset time window based on the dynamic decay model of silencing activity. Within the prediction time window, the algorithm comprehensively evaluates the cumulative impact of the current control action on the future silencing efficiency, fruit health and environmental energy consumption, selects the control action sequence that maximizes the comprehensive benefits within the entire prediction window, and executes the first action of the sequence as the optimal control action at the current moment, so as to achieve proactive response and preventive control to environmental disturbances.

[0035] The online adaptive update phase includes an update protection mechanism: before each parameter fine-tuning of the initial network model, the distribution similarity between the current batch of real-time data and the offline training dataset is first calculated; when the distribution similarity is lower than a preset threshold, the online update of model parameters is paused, and a rule-based control strategy is adopted as a backup scheme to output control actions. A rapid data acquisition process is initiated based on the current batch of data. After the amount of acquired data meets the requirements, the distribution similarity is re-evaluated to determine whether to resume model updates. At the same time, during the model update process, an elastic weight solidification algorithm is introduced to impose a high update penalty on network parameters that affect important features of historical data to prevent catastrophic forgetting.

[0036] The workflow of this invention: I. Multimodal Data Acquisition Immediately after VIGS inoculation of the pepper fruits, a multimodal sensor array is deployed. This system continuously collects environmental factor data from the fruit microenvironment at a preset first frequency, including at least temperature, humidity, light intensity, and... Concentration; simultaneously, physiological state data of the fruit are continuously collected at a preset second frequency, including at least the water potential of the fruit tissue, cell membrane permeability, and the relative expression level of target gene mRNA measured by minimally invasive sampling combined with quantitative PCR technology.

[0037] II. Construction of State Quantization and Decay Model Based on the relative expression levels of target gene mRNA collected in real time in the first step, the system calculates the VIGS silencing efficiency index at the current moment. Then, using time as the independent variable and the silencing efficiency index as the dependent variable, data fitting is performed to construct a dynamic decay model of silencing activity for individual fruits. Using this model, the system can predict the remaining maintenance time required for the silencing efficiency index to decay to a preset effectiveness threshold, achieving quantitative assessment and trend prediction of VIGS activity status.

[0038] III. Feedforward-Feedback Composite Control The environmental factor data collected in the first step, the water potential of the fruit tissue, the cell membrane permeability, and the remaining maintenance time predicted in the second step are used as input variables and fed into a pre-trained state-value network. This network is built based on a deep reinforcement learning model, and its optimization objective is to maximize the duration for which the silencing efficiency index is higher than a preset threshold while minimizing the energy consumption generated by environmental regulation. After calculation, the network outputs the optimal regulation action at the current moment. This action can be adjusting the set values ​​of environmental factors (such as temperature and humidity) or starting / stopping the application of exogenous hormones.

[0039] IV. Personalized Control Implementation and Closed-Loop Feedback Based on the optimal regulatory actions generated in the third step, the system generates specific control commands and sends them to environmental regulation actuators (such as humidifiers, heaters, and supplemental lighting) to dynamically and individually adjust the microenvironment of the fruit. Simultaneously, the system repeats steps S1 to S4 at a preset third frequency, forming a continuous closed-loop regulatory process. This process will continue indefinitely until the remaining maintenance time predicted in the second step decays to zero, thus maintaining highly efficient gene silencing activity throughout the entire VIGS experimental cycle.

[0040] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for maintaining VIGS activity in chili pepper fruits based on data analysis, characterized in that, Includes the following steps: S1. Multimodal Data Acquisition: After VIGS inoculation of pepper fruits, a multimodal sensor array is deployed to continuously acquire environmental factor data of the microenvironment where the fruit is located at a preset first frequency, and to continuously acquire physiological state data of the fruit at a preset second frequency. The environmental factor data includes at least temperature, humidity, light intensity, and... The concentration, the physiological state data includes at least the water potential of fruit tissue, cell membrane permeability, and the relative expression level of target gene mRNA determined by minimally invasive sampling combined with quantitative PCR technology; S2. State Quantification and Decay Model Construction: Based on the real-time acquisition of the relative expression level of target gene mRNA, the VIGS silencing efficiency index at the current moment is calculated. With time as the independent variable and the silencing efficiency index as the dependent variable, a dynamic decay model of the silencing activity of individual fruits is constructed by fitting and predicting the remaining maintenance time required for the silencing efficiency index to decay to a preset threshold. S3. Feedforward-feedback composite control: The environmental factor data, fruit tissue water potential, cell membrane permeability, and remaining maintenance time are used as input variables and input into a pre-trained state-action value network. The state-action value network is constructed based on a deep reinforcement learning model and aims to maximize the duration for which the silencing efficiency index is higher than the preset threshold and minimize the energy consumption of environmental regulation. The state-action value network outputs the optimal regulation action at the current moment. The regulation action includes adjusting the environmental factor setpoint or starting / stopping the application of exogenous hormones. S4. Personalized control execution: Based on the optimal control action, a control command is generated and sent to the environmental control execution mechanism to dynamically adjust the fruit microenvironment. At the same time, steps S1 to S4 are repeated at a preset third frequency to form a closed-loop control until the remaining maintenance time is zero.

2. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 1, characterized in that, The specific process of constructing the dynamic decay model of the silent activity of individual fruits in step S2 is as follows: Based on a nonlinear mixed-effects model, the change in VIGS silencing efficiency over time is decomposed into a fixed-effects component and a random-effects component. The fixed-effects component describes the average decay of silencing efficiency over time at the population level, while the random-effects component describes the degree to which individual fruits deviate from the population average due to differences in inoculation and physiological state. Using a Bayesian parameter inference method, the random-effects parameters of individual fruits are updated online using real-time collected data on the relative expression levels of target gene mRNAs. This generates a personalized dynamic decay model of silencing activity in real time, and the remaining maintenance time is predicted based on this model.

3. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 1, characterized in that, The training process of the state-action value network in step S3 includes: A computational simulation environment was constructed to simulate the VIGS silencing process under different environmental disturbances and fruit physiological states. In this simulation environment, a dynamic decay model of silencing activity driven by historical experimental data was used as the core. A state space, action space, and reward function were defined. The reward function consisted of a silencing efficiency maintenance reward term and an energy consumption penalty term. Through reinforcement learning algorithms, the agent was driven to conduct massive trial-and-error learning in the simulation environment to maximize the cumulative reward, thereby obtaining the state-action value network that selects the optimal regulatory action under different states. After training, the state-action value network was deployed to the actual regulatory system for online decision-making.

4. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 1, characterized in that, In step S1, a near-infrared spectral sensor is used to replace the minimally invasive sampling combined with quantitative PCR technology. Characteristic spectral data of the fruit are collected in real time in a non-contact manner. Based on a pre-established spectral-silencing efficiency correlation model, the characteristic spectral data is converted into an equivalent target gene silencing efficiency index.

5. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 1, characterized in that, The optimal control action in step S3 is generated based on a multi-objective optimization algorithm. The multi-objective optimization algorithm takes maintaining the silencing efficiency index within a preset qualified range, maintaining the fruit tissue water potential within a preset healthy range, and minimizing the energy consumption of environmental control as the collaborative optimization objectives. When the silencing efficiency index is higher than the upper limit of the preset qualified range, the control action prioritizes reducing the intensity of environmental control rather than starting hormone spraying, so as to avoid causing excessive stress to the fruit.

6. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 1, characterized in that, The state-action value network in step S3 has two training stages: offline pre-training and online adaptive update. In the offline pre-training stage, an initial network model is trained using experimental data containing multiple chili varieties and historical environmental conditions. During the online adaptive update phase, when the system is deployed on a new fruit population, the silencing efficiency data and environmental response data of the current batch are collected in real time. A meta-learning algorithm is used to quickly fine-tune some parameters of the initial network model to adapt it to the dynamic characteristics of the current variety and environment. The updated model is used for subsequent regulatory decisions.

7. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 4, characterized in that, Step S1 also includes a self-calibration step for the spectral sensor: a reference standard with stable spectral reflectance characteristics is fixedly set next to the fruit sample. Before each collection of fruit spectral data, the spectral signal of the reference standard is first collected, the deviation between the reference standard and the standard reference spectrum is calculated, and the subsequent collected fruit characteristic spectral data is corrected in real time based on the deviation to eliminate systematic errors introduced by sensor drift and environmental fluctuations. At the same time, the spectral-silence efficiency correlation model is constructed using an ensemble learning algorithm, which integrates the outputs of multiple sub-models of different bands. When a certain band signal is disturbed, its weight is automatically reduced to ensure the robustness of the silence efficiency index estimation.

8. The method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 5, characterized in that, The multi-objective optimization algorithm is a model predictive control algorithm. This algorithm not only optimizes based on the current state, but also makes rolling predictions of the fruit state within a preset time window based on the dynamic decay model of silencing activity. Within the prediction time window, the algorithm comprehensively evaluates the cumulative impact of the current control action on future silencing efficiency, fruit health and environmental energy consumption, selects the control action sequence that maximizes the comprehensive benefits within the entire prediction window, and executes the first action of the sequence as the optimal control action at the current moment, thereby achieving proactive response and preventive control to environmental disturbances.

9. A method for maintaining VIGS activity in chili pepper fruits based on data analysis according to claim 6, characterized in that, The online adaptive update phase includes an update protection mechanism: before each parameter fine-tuning of the initial network model, the distribution similarity between the current batch of real-time data and the offline training dataset is first calculated; when the distribution similarity is lower than a preset threshold, the online update of model parameters is paused, and a rule-based control strategy is adopted as a backup scheme to output control actions. A rapid data acquisition process is initiated based on the current batch of data. After the amount of acquired data meets the requirements, the distribution similarity is re-evaluated to determine whether to resume model updates. At the same time, during the model update process, an elastic weight solidification algorithm is introduced to apply a high update penalty to network parameters that affect important features of historical data to prevent catastrophic forgetting.

Images