Regulation and control method for enhancing photosynthetic efficiency through hormone signal transduction between scions of grafted rootstock

Abstract

The invention discloses a regulation and control method for enhancing photosynthetic efficiency through hormone signal transduction between grafted rootstocks and scions, and particularly relates to the field of photosynthetic efficiency regulation and control. The regulation and control method comprises the steps that hormone release data of rootstocks, photosynthetic response data of scions and corresponding time labels are obtained, and according to the time difference between hormone change and photosynthetic response, the photosynthetic response data of the scions are obtained; constructing a state sequence with a delay tag; and defining an action set according to the delay characteristics marked in the state sequence. By constructing a state transition structure of delayed response perception and a regulation strategy generation mechanism based on a time recursion feature, an onset time point of a hormone signal and a feedback response track of a scion are captured, and accurate identification and intervention opportunity optimization of a regulation window in a non-dominant coupling chain are realized; therefore, the problems that an existing model cannot effectively model a time evolution structure, strategy execution is distorted, and improvement of photosynthetic efficiency is difficult to stably capture are solved.

Description

Technical Field

[0001] This invention relates to the field of photosynthetic efficiency regulation technology, and more specifically, to a method for regulating hormone signal transduction between grafted rootstock and scion to enhance photosynthetic efficiency. Background Technology

[0002] In current deep learning-based plant physiological regulation technologies, hormone signal modeling mostly adopts static or quasi-static state-action mapping methods, ignoring the complex temporal evolution mechanism between rootstock and scion within the plant. Especially in grafting systems, hormone conduction is not instantaneous. After crossing the graft union, it exhibits significant dynamic delay effects and nonlinear interaction characteristics due to differences in inter-tissue physiological permeability, signal response thresholds, and feedback regulation mechanisms. Furthermore, this signal flow is not a unidirectional linear transmission but involves multi-level feedback and periodic interactions, and is highly dependent on the current physiological state of the plant and disturbances in the external environment.

[0003] Therefore, if static feature extraction or state estimation within a short window is still used, it is easy to miss the key effective nodes of the rootstock signal and the actual response trajectory of the scion, which will lead to judgment lag and strategy deviation during the regulation process. The model will be unable to accurately define "when to intervene" and "what kind of feedback is an effective response", which will ultimately result in the difficulty of the system to stably capture the photosynthetic efficiency improvement effect of hormone regulation.

[0004] Based on the above analysis, a core problem can be summarized: existing regulatory models lack the ability to deeply model the time-recursive structure of hormone signal transduction in rootstock-scion, making it difficult to identify the non-obvious coupling relationship of "signal activation-feedback response" in the physiological response chain, resulting in distorted regulatory strategies and missed photosynthetic optimization windows. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency. By constructing a state transition structure with delayed response perception and a regulation strategy generation mechanism based on time recursion characteristics, the method captures the onset time of hormone signals and the feedback response trajectory of the scion, thereby achieving accurate identification of the regulation window in the non-obvious coupling chain and optimization of intervention timing. This solves the problems of existing models being unable to effectively model the time evolution structure, strategy execution distortion, and difficulty in stably capturing the improvement in photosynthetic efficiency.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for regulating hormone signal transduction between grafted rootstock and scion to enhance photosynthetic efficiency, comprising:

[0007] S1. Obtain hormone release data from rootstock, photosynthetic response data from scion, and corresponding time tags. Based on the time difference between hormone changes and photosynthetic response, construct a state sequence with delay tags.

[0008] S2. Based on the delay features marked in the state sequence, define an action set. Each action in the action set includes the type of hormone release, the release time, and the duration of action. The action set is used to form the action selection space required for the regulatory strategy.

[0009] S3. Construct a Q-network model containing recursive units, with state sequences as input and action sets as output. By recording the combined trajectories of historical states and actions, update the expected regulatory value of each action in different states.

[0010] S4. During training, the error is calculated for the photosynthetic response time points corresponding to the state and action, the actual response delay is identified, and the actual response delay is fed back to adjust the evaluation value of the relevant action in the Q network.

[0011] S5. Based on the current rootstock status and the trained Q-network structure, select the corresponding optimal hormone release action and its execution time as a regulation scheme for regulating the photosynthetic behavior of the scion, and implement hormone regulation according to the regulation scheme.

[0012] In a preferred embodiment, in S1, hormone release data of the rootstock during the target period is acquired, and the hormone type, hormone concentration value and corresponding time information at each moment are recorded to generate a hormone data sequence arranged in chronological order.

[0013] Acquire photosynthetic response data of the scion within the target period, record the photosynthetic response parameters and their time information at each moment, and generate a photosynthetic data sequence corresponding to the hormone data sequence on the time axis;

[0014] Hormone data sequences and photosynthetic data sequences are time-synchronized, and data with the same time label are aligned to form initial data pair sequences based on time pairing.

[0015] In a preferred embodiment, in S1, the time difference between the hormone release time point and the photosynthetic response time point in each data pair is calculated, the response delay value of the data pair is extracted, and a delayed data sequence containing time offset information is generated.

[0016] Based on the magnitude of each response delay value in the delayed data sequence, the corresponding data pairs are labeled with delay tags. The delay tags are used to identify whether the photosynthetic response can be attributed to the current hormone release behavior.

[0017] Hormone type, hormone concentration, photosynthetic response parameters, time label, and delay label are uniformly encoded as state elements. The state elements are then combined in chronological order to construct a state sequence with delay labels, which serves as the temporal structure data for subsequent regulatory model input.

[0018] In a preferred embodiment, in S2, by performing a parsing operation on the state elements containing delay labels in the state sequence, data on hormone release type, hormone release time point, photosynthetic response time point, and the time interval between the two are extracted from each state element, and a set of state parameters containing delay information is output.

[0019] The set of state parameters containing delay information is segmented and classified according to the time interval, and the number of photosynthetic responses and the average response intensity of each hormone type in each time interval are statistically analyzed within that interval, and a set of hormone response structures with interval statistical indicators is output.

[0020] By performing response interval feature extraction on the hormone response structure set, the average hormone release offset and duration of effect of each hormone type in its respective response interval are identified, and the results are output as the standard release time and standard effect time.

[0021] In a preferred embodiment, in S2, the hormone release type, standard release time, and standard action time are combined into action tuples, and the average response intensity score corresponding to the response structure set is added to each action tuple to construct a set of action tuples containing score labels.

[0022] By comparing the response intensity score and time interval coverage ratio of each action tuple in the action tuple set with thresholds, action tuples that simultaneously meet the conditions of having a response intensity higher than the preset response intensity threshold and a coverage ratio higher than the preset coverage ratio threshold are selected, and all action tuples that meet the conditions are combined to form a set of strategy candidate actions.

[0023] Output all action tuples in the policy candidate action set as the action set in the control model.

[0024] In a preferred embodiment, in S3, each state element s in the constructed state sequence is... t The data are sequentially input into a state representation encoding network, where multidimensional temporal embedding and sensor feature extraction operations are performed to construct a state embedding vector h. t =φ(s) t ), where the state element s t This represents the state unit at time t, which includes the hormone release value, the photosynthetic response value, and the time delay. Let d represent the state representation vector of dimension d; φ(·) is the state embedding function;

[0025] The states are embedded into the vector sequence {h1,h2,...,h...} TThe data is input sequentially into a memory recursive unit containing a gating structure to construct a recursive state trajectory z that incorporates historical dependencies. t ,Right now:

[0026] z t =GRU(h1,h2,...,h t )

[0027] in The vector represents the state trajectory with delayed memory, T represents the total number of time steps used for training, and GRU stands for Gated Recurrent Unit.

[0028] For each action tuple a in the candidate action set i The corresponding action vector u is obtained by performing structured parameter transformation through an action feature encoder. i =ψ(a i ), where a i Includes hormone type, hormone release time point, and duration; Let ψ(·) be the action representation vector; ψ(·) be the action encoding function.

[0029] The current state trajectory vector z t With all candidate action vectors u i The concatenated data is then input into the action scoring network f0, and the network outputs the moderating value estimate for each action in the current state.

[0030]

[0031] in Indicates action a i In state z t The expected regulatory value under the current circumstances This represents a vector concatenation operation, where f0 is a neural network function with parameter θ.

[0032] In a preferred embodiment, S3 further includes: for each executed state-action-feedback record (z t ,a t ,r t ,z t+1 ), to build value renewal goals:

[0033]

[0034] The squared error between the predicted value and the target value is calculated as the loss function:

[0035]

[0036] Where: r t Indicates action a in the current state tThe obtained photosynthetic response feedback; action a t Indicates that in state z t The action to be performed is selected below; z t+1 This represents the state trajectory vector of the system at the next moment after the action is performed; γ∈0,1 is the discount factor. Represents the complete set of actions; a′ represents the candidate action for evaluating the maximum Q value in the next state; Q(z) t ,a t ) indicates that in state z t Next, execute action a t The expected regulatory value that can be obtained; Indicates that in state z t+1 The action with the highest expected value among all candidate actions is evaluated; T represents the total number of time steps used for training; y t Indicates based on the current feedback r t The target control value is calculated based on the maximum action value of the future state.

[0037] Using optimizers based on loss functions Backpropagation and gradient updates are performed on the parameters θ of the scoring network, enabling the recursive network to gradually optimize the hormone regulation strategy in the state-action space.

[0038] In a preferred embodiment, in S4, the action execution time t in each state trajectory is determined. a The actual peak response time t in the corresponding photosynthetic response data r Perform time difference calculations to obtain the actual response delay value Δt = t for each action in the state trajectory. r -t a ;

[0039] The actual response delay value is matched with the control prediction window corresponding to the action trigger time to determine whether the current control behavior exceeds the policy prediction range, and the delay deviation label matrix is ​​output.

[0040] Based on the delay bias label matrix, in the original feedback value r t Apply a time-sensitive function adjustment to construct a delay-sensitive feedback value. Where γ∈(0,1) is the delay discount factor;

[0041] By z t a t and Combining to construct new training triples Training Triples Objectives used to regulate the value function;

[0042] Using the training triples Perform target regression calculation of the state-action mapping function in the deep Q network to update the expected regulatory value estimate of each action under different states;

[0043] During training, the distribution of delayed behavior at consecutive time steps is recorded, and a dynamic response window scheduling strategy is constructed based on the cumulative trend of delay bias to correct the time tolerance interval for future action selection in the Q network.

[0044] In a preferred embodiment, in S5, by inputting the current state of the rootstock into the trained delayed-aware Q-network model, the action regulation value distribution corresponding to the current state is extracted, wherein the state includes the rootstock hormone level, environmental parameters and the residual effect of the previous action.

[0045] The extracted action regulation value distribution is dynamically mapped to a preset value evaluation function, wherein the value evaluation function is dynamically updated based on the historical state-action-reward sequence, and the regulation value of each action is normalized during the mapping process.

[0046] All hormone-releasing actions in the action set are sorted according to the normalized regulatory value, and the effect stability confidence of each hormone-releasing action in the current state is calculated. The confidence is obtained based on the response delay variance corresponding to similar states in the training phase.

[0047] From the ranking results, select the hormone release actions with the greatest regulatory value and confidence level above the preset threshold, and extract their corresponding hormone type, release time and duration of action to construct a subset of state-associated hormone regulatory actions.

[0048] In a preferred embodiment, in S5, each action in the candidate subset of hormone regulation actions is compiled together with the current hormone reserve data and recent environmental prediction data into an input data packet.

[0049] The input data packet is fed into the existing strategy generation module, where it is determined whether each hormone action meets the current hormone reserve requirements and whether it is suitable for the predicted environmental conditions.

[0050] Select the action plan that meets the requirements as the current control plan, and finally output a set of target control plans that meet the requirements of resource constraints and environmental adaptability;

[0051] The hormone release control device, according to the hormone release type, release time and release dose specified in the regulation scheme, completes the quantitative hormone release operation on the rootstock, thereby achieving target-oriented regulation of the photosynthetic behavior of the scion.

[0052] The technical effects and advantages of this invention are as follows:

[0053] 1. This scheme constructs a delayed-sensing Q-network to dynamically model the "signal onset - feedback response" delay relationship in the hormone signal transduction process between rootstock and scion. This overcomes the problem in existing technologies that cannot accurately identify the physiological onset time and response window of the signal, thereby avoiding deviation of the regulatory strategy and missed execution opportunities, and improving the timeliness and response accuracy of the photosynthetic efficiency optimization strategy under hormone regulation.

[0054] 2. This scheme introduces a segmentation and classification mechanism for hormone response behavior in the action set construction stage. By identifying key response boundaries and time markers in the state evolution process, the strategy model can distinguish the role patterns of different hormone regulatory behaviors in the scion response.

[0055] 3. This scheme utilizes a Q-network with a recursive memory structure to perform deep learning on the potential dynamic associations in the historical state sequence, which strengthens the policy's understanding of the causal structure of the "hormone release - photosynthetic response" chain in the time-series pattern and avoids the misjudgment problem caused by the lack of state evolution memory in the static policy model.

[0056] 4. This solution introduces delay deviation labels and dynamic response windows to achieve timely correction and window reconstruction of strategy behavior under different physiological stages and environmental disturbances.

[0057] 5. During the strategy execution phase, this plan introduces a dual evaluation mechanism of regulatory value and effect confidence, and dynamically generates regulatory outputs in combination with resource constraints, forming a more targeted regulatory plan in terms of the intensity, type and timing of hormone release, thus avoiding resource waste and over-intervention. Attached Figure Description

[0058] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0060] Refer to the instruction manual appendix Figure 1 A method for regulating hormone signal transduction between grafted rootstock and scion to enhance photosynthetic efficiency according to an embodiment of the present invention includes:

[0061] S1. Obtain hormone release data from rootstock, photosynthetic response data from scion, and corresponding time tags. Based on the time difference between hormone changes and photosynthetic response, construct a state sequence with delay tags.

[0062] S2. Based on the delay features marked in the state sequence, define an action set. Each action in the action set includes the type of hormone release, the release time, and the duration of action. The action set is used to form the action selection space required for the regulatory strategy.

[0063] S3. Construct a Q-network model containing recursive units, with state sequences as input and action sets as output. By recording the combined trajectories of historical states and actions, update the expected regulatory value of each action under different states to form a stable evaluation of delayed regulatory behavior.

[0064] S4. During training, the error is calculated for the photosynthetic response time points corresponding to the state and action, the actual response delay is identified, and the actual response delay is fed back to adjust the evaluation value of the relevant action in the Q network to ensure that the strategy is adaptable to the response delay.

[0065] S5. Based on the current rootstock status and the trained Q-network structure, select the corresponding optimal hormone release action and its execution time as a regulation scheme for regulating the photosynthetic behavior of the scion, and implement hormone regulation according to the regulation scheme.

[0066] In S1, hormone release data of the rootstock during the target period is acquired, and the hormone type, hormone concentration value and corresponding time information at each moment are recorded to generate a hormone data sequence arranged in chronological order.

[0067] Acquire photosynthetic response data of the scion within the target period, record the photosynthetic response parameters and their time information at each moment, and generate a photosynthetic data sequence corresponding to the hormone data sequence on the time axis;

[0068] Hormone data sequences and photosynthetic data sequences are time-synchronized, and data with the same time label are aligned to form initial data pair sequences based on time pairing.

[0069] In S1, the time difference between the hormone release time point and the photosynthetic response time point in each data pair is calculated, the response delay value of the data pair is extracted, and a delayed data sequence containing time offset information is generated.

[0070] Based on the magnitude of each response delay value in the delayed data sequence, the corresponding data pairs are labeled with delay tags. The delay tags are used to identify whether the photosynthetic response can be attributed to the current hormone release behavior.

[0071] Hormone type, hormone concentration, photosynthetic response parameters, time label, and delay label are uniformly encoded as state elements. The state elements are then combined in chronological order to construct a state sequence with delay labels, which serves as the temporal structure data for subsequent regulatory model input.

[0072] In S2, by performing a parsing operation on the state elements containing delay labels in the state sequence, data on hormone release type, hormone release time point, photosynthetic response time point, and the time interval between the two are extracted from each state element, and a set of state parameters containing delay information is output.

[0073] The set of state parameters containing delay information is segmented and classified according to the time interval, and the number of photosynthetic responses and the average response intensity of each hormone type in each time interval are statistically analyzed within that interval, and a set of hormone response structures with interval statistical indicators is output.

[0074] By performing response interval feature extraction on the hormone response structure set, the average hormone release offset and duration of effect of each hormone type in its respective response interval are identified, and the results are output as the standard release time and standard effect time.

[0075] In S2, hormone release type, standard release time and standard action time are combined into action tuples, and the average response intensity score corresponding to the response structure set is added to each action tuple to construct a set of action tuples with score annotations.

[0076] By comparing the response intensity score and time interval coverage ratio of each action tuple in the action tuple set with thresholds, action tuples that simultaneously meet the conditions of having a response intensity higher than the preset response intensity threshold and a coverage ratio higher than the preset coverage ratio threshold are selected, and all action tuples that meet the conditions are combined to form a set of strategy candidate actions.

[0077] All action tuples in the policy candidate action set are output as the action set in the control model, and this action set is defined as the action selection space for use by the subsequent policy training module.

[0078] In S3, each state element s in the completed state sequence will be... t The data are sequentially input into a state representation encoding network, where multidimensional temporal embedding and sensor feature extraction operations are performed to construct a state embedding vector h. t =φ(s) t ), where the state element s t This represents the state unit at time t, which includes the hormone release value, the photosynthetic response value, and the time delay. Let d represent the state representation vector of dimension d; φ(·) is the state embedding function;

[0079] The states are embedded into the vector sequence {h1,h2,...,h...} T The data is input sequentially into a memory recursive unit containing a gating structure to construct a recursive state trajectory z that incorporates historical dependencies. t ,Right now:

[0080] z t =GRU(h1,h2,...,h t )

[0081] in is the state trajectory vector with delayed memory, T represents the total number of time steps used for training; GRU is a gated recurrent unit, which is a recurrent neural network structure used to retain long-term dependency information and reduce the gradient vanishing problem in sequential data through reset gate and update gate mechanisms;

[0082] For each action tuple a in the candidate action set i The corresponding action vector u is obtained by performing structured parameter transformation through an action feature encoder. i =ψ(a i ), where a i Includes hormone type, hormone release time point, and duration; Let ψ(·) be the action representation vector; ψ(·) be the action encoding function.

[0083] The current state trajectory vector z t With all candidate action vectors u i The concatenated data is then input into the action scoring network f0, and the network outputs the moderating value estimate for each action in the current state.

[0084]

[0085] in Indicates action a i In state z t The expected regulatory value is given by ⊕, which represents the vector concatenation operation, and f0 is a neural network function with parameter θ.

[0086] S3 also includes: a state-action-feedback log for each executed action (z t ,a t ,r t ,z t+1 ), to build value renewal goals:

[0087]

[0088] The squared error between the predicted value and the target value is calculated as the loss function:

[0089]

[0090] Where: r t Indicates action a in the current state t The obtained photosynthetic response feedback; action a t Indicates that in state z t The action to be performed is selected below; z t+1 This represents the state trajectory vector of the system at the next moment after the action is performed; γ∈0,1 is the discount factor. Represents the complete set of actions; a′ represents the candidate action for evaluating the maximum Q value in the next state; Q(z) t ,a t ) indicates that in state z t Next, execute action a t The expected regulatory value that can be obtained; Indicates that in state z t+1 The action with the highest expected value among all candidate actions is evaluated; T represents the total number of time steps used for training; y t Indicates based on the current feedback r t The target control value, calculated from the maximum action value of the future state, is used to guide the value update of the Q network;

[0091] Using optimizers based on loss functions Backpropagation and gradient updates are performed on the parameters θ of the scoring network, enabling the recursive network to gradually optimize the hormone regulation strategy in the state-action space, forming a policy network with stable response value estimation capabilities under different delay states.

[0092] In S4, the execution time t of each action in the state trajectory is determined. a The actual peak response time t in the corresponding photosynthetic response data r Perform time difference calculations to obtain the actual response delay value Δt = t for each action in the state trajectory. r -t a ;

[0093] The actual response delay value is matched with the control prediction window corresponding to the action trigger time to determine whether the current control behavior exceeds the policy prediction range, and the delay deviation label matrix is ​​output.

[0094] Based on the delay bias label matrix, in the original feedback value r t Apply a time-sensitive function adjustment to construct a delay-sensitive feedback value. Where γ∈(0,1) is the delay discount factor;

[0095] By z t a t and Combining to construct new training triples Training Triples Objectives used to regulate the value function;

[0096] Using the training triples Perform target regression calculation of the state-action mapping function in the deep Q network to update the expected regulatory value estimate of each action under different states;

[0097] During training, the distribution of delayed behavior at consecutive time steps is recorded, and a dynamic response window scheduling strategy is constructed based on the cumulative trend of delay bias to correct the time tolerance interval for future action selection in the Q network.

[0098] In S5, by inputting the current state of the rootstock into the trained delayed-aware Q-network model, the action regulation value distribution corresponding to the current state is extracted. The state includes the rootstock hormone level, environmental parameters, and residual effect of the previous action.

[0099] The extracted action regulation value distribution is dynamically mapped to a preset value evaluation function, wherein the value evaluation function is dynamically updated based on the historical state-action-reward sequence, and the regulation value of each action is normalized during the mapping process.

[0100] All hormone-releasing actions in the action set are sorted according to the normalized regulatory value, and the effect stability confidence of each hormone-releasing action in the current state is calculated. The confidence is obtained based on the response delay variance corresponding to similar states in the training phase.

[0101] From the ranking results, select the hormone release actions with the greatest regulatory value and confidence level above the preset threshold, and extract their corresponding hormone type, release time and duration of action to construct a subset of state-associated hormone regulatory actions.

[0102] In S5, each action in the candidate subset of hormone regulation actions is combined with current hormone reserve data and recent environmental prediction data into an input data package;

[0103] The input data packet is fed into the existing strategy generation module, where it is determined whether each hormone action meets the current hormone reserve requirements and whether it is suitable for the predicted environmental conditions.

[0104] Select the most suitable action plan from the actions that meet the requirements for current regulation, and finally output a set of target regulation plans that meet resource constraints and environmental adaptability.

[0105] The hormone release control device, according to the hormone release type, release time and release dose specified in the regulation scheme, completes the quantitative hormone release operation on the rootstock, thereby achieving target-oriented regulation of the photosynthetic behavior of the scion.

[0106] It should be noted that this scheme is based on a systematic analysis of the defects of the current plant regulation system in terms of delayed response recognition, feedback regulation ability and resource regulation efficiency, especially the characteristic recognition and utilization mechanism of the significant time lag between rootstock hormone release behavior and scion photosynthetic response in heterogeneous plant grafting structures.

[0107] The entire solution is built step by step through the steps S1 to S5 mentioned above. Its formation process strictly follows the closed-loop chain structure of state data acquisition, delay feature annotation, action set construction, policy network training, feedback correction and final control execution.

[0108] Specifically, in S1, by acquiring rootstock hormone release data and scion photosynthetic response data, and analyzing and processing the correspondence between the two on the time axis, a state sequence with delay labels is constructed. This process is the core of the data foundation construction for the entire scheme. In S2, delay information is extracted from the state sequence and the response behavior is segmented and classified. The interval characteristics and regulatory boundaries of hormone response are extracted, and an action set with scores and time-effect labels is constructed as the input space for subsequent strategy learning, ensuring the pertinence and selectivity of the regulatory strategy.

[0109] In S3, the action set is input into a Q-network with a recursive structure. Its sequence modeling capability is used to predict the execution value of historical state-action combinations under delayed conditions. During training, the state trajectory is continuously recorded, enabling the model to maintain policy coherence and prediction accuracy when facing unstable responses. S4 further enhances the policy network's ability to perceive and adapt to delay changes. By analyzing the matching degree between the actual response time and the policy prediction behavior, a delay bias label is constructed, and the action feedback value is corrected accordingly. A time-sensitive mechanism is introduced to improve the system's feedback accuracy under dynamic conditions. A dynamic response window is constructed through the historical trend of delay distribution, thereby dynamically adjusting the time tolerance range of future regulatory behaviors, significantly enhancing the spatiotemporal adaptability of the policy model.

[0110] In S5, the scheme enters the final execution logic stage. The current rootstock state is input into the trained delayed-sensing Q-network, which outputs the regulatory value and effect confidence of each hormone release action in the current state. Based on resource accessibility and environmental prediction data, the optimal regulation scheme is dynamically selected in the strategy generation module. Then, the selected hormone type, release time and dosage are precisely executed through the control device to achieve target-oriented regulation of the photosynthetic efficiency of the scion.

[0111] The reason for choosing this design path is based on the overall judgment that the plant signal transduction process is subject to multiple lags, strong external disturbances, and high resource allocation requirements. This solution upgrades the traditional static hormone regulation strategy into a dynamic intelligent regulation system with state tracking and delay perception capabilities by integrating three mechanisms: delay modeling, strategy selection, and feedback correction. This improves the adaptability, execution stability, and regulation accuracy of the regulation behavior of heterogeneous grafted plants in actual agricultural scenarios, and provides an implementable and scalable technical path for efficient regulation under complex plant structures.

[0112] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. Methods for regulating hormone signal transduction between grafted rootstock and scion to enhance photosynthetic efficiency, including: S1. Obtain hormone release data from rootstock, photosynthetic response data from scion, and corresponding time tags. Based on the time difference between hormone changes and photosynthetic response, construct a state sequence with delay tags. Its features are: S2. Based on the delay features marked in the state sequence, define an action set. Each action in the action set includes the type of hormone release, the release time, and the duration of action. The action set is used to form the action selection space required for the regulatory strategy. S3. Construct a Q-network model containing recursive units, with state sequences as input and action sets as output. By recording the combined trajectories of historical states and actions, update the expected regulatory value of each action in different states. S4. During training, the error is calculated for the photosynthetic response time points corresponding to the state and action, the actual response delay is identified, and the actual response delay is fed back to adjust the evaluation value of the relevant action in the Q network. S5. Based on the current rootstock status and the trained Q-network structure, select the corresponding optimal hormone release action and its execution time as a regulation scheme for regulating the photosynthetic behavior of the scion, and implement hormone regulation according to the regulation scheme.

2. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 1, characterized in that: In S1, hormone release data of the rootstock during the target period is acquired, and the hormone type, hormone concentration value and corresponding time information at each moment are recorded to generate a hormone data sequence arranged in chronological order. Acquire photosynthetic response data of the scion within the target period, record the photosynthetic response parameters and their time information at each moment, and generate a photosynthetic data sequence corresponding to the hormone data sequence on the time axis; Hormone data sequences and photosynthetic data sequences are time-synchronized, and data with the same time label are aligned to form initial data pair sequences based on time pairing.

3. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 2, characterized in that: In S1, the time difference between the hormone release time point and the photosynthetic response time point in each data pair is calculated, the response delay value of the data pair is extracted, and a delayed data sequence containing time offset information is generated. Based on the magnitude of each response delay value in the delayed data sequence, the corresponding data pairs are labeled with delay tags. The delay tags are used to identify whether the photosynthetic response can be attributed to the current hormone release behavior. Hormone type, hormone concentration, photosynthetic response parameters, time label, and delay label are uniformly encoded as state elements. The state elements are then combined in chronological order to construct a state sequence with delay labels, which serves as the temporal structure data for subsequent regulatory model input.

4. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 3, characterized in that: In S2, by performing a parsing operation on the state elements containing delay labels in the state sequence, data on hormone release type, hormone release time point, photosynthetic response time point, and the time interval between the two are extracted from each state element, and a set of state parameters containing delay information is output. The set of state parameters containing delay information is segmented and classified according to the time interval, and the number of photosynthetic responses and the average response intensity of each hormone type in each time interval are statistically analyzed within that interval, and a set of hormone response structures with interval statistical indicators is output. By performing response interval feature extraction on the hormone response structure set, the average hormone release offset and duration of effect of each hormone type in its respective response interval are identified, and the results are output as the standard release time and standard effect time.

5. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 4, characterized in that: In S2, hormone release type, standard release time and standard action time are combined into action tuples, and the average response intensity score corresponding to the response structure set is added to each action tuple to construct a set of action tuples with score annotations. By comparing the response intensity score and time interval coverage ratio of each action tuple in the action tuple set with thresholds, action tuples that simultaneously meet the conditions of having a response intensity higher than the preset response intensity threshold and a coverage ratio higher than the preset coverage ratio threshold are selected, and all action tuples that meet the conditions are combined to form a set of strategy candidate actions. Output all action tuples in the policy candidate action set as the action set in the control model.

6. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 5, characterized in that: In S3, each state element s in the completed state sequence will be... t The data are sequentially input into a state representation encoding network, where multidimensional temporal embedding and sensor feature extraction operations are performed to construct a state embedding vector h. t =φ(s) t ), where the state element s t This represents the state unit at time t, which includes the hormone release value, the photosynthetic response value, and the time delay. Let d represent the state representation vector of dimension d; φ(·) is the state embedding function; The states are embedded into the vector sequence {h1,h2,...,h...} T The data is input sequentially into a memory recursive unit containing a gating structure to construct a recursive state trajectory z that incorporates historical dependencies. t ,Right now: from t =GROUP(h1,h2,...,h t ) in The vector represents the state trajectory with delayed memory, T represents the total number of time steps used for training, and GRU stands for Gated Recurrent Unit. For each action tuple a in the candidate action set i The corresponding action vector u is obtained by performing structured parameter transformation through an action feature encoder. i =ψ(a i ), where a i Includes hormone type, hormone release time point, and duration; Let ψ(·) be the action representation vector; ψ(·) be the action encoding function. The current state trajectory vector z t With all candidate action vectors u i The concatenated data is then input into the action scoring network f0, and the network outputs the moderating value estimate for each action in the current state. in Indicates action a i In state z t The expected regulatory value under the current circumstances This represents a vector concatenation operation, where f0 is a neural network function with parameter θ.

7. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 6, characterized in that: S3 also includes: a state-action-feedback log for each executed action (z t ,a t ,r t ,z t+1 ), to build value renewal goals: The squared error between the predicted value and the target value is calculated as the loss function: Where: r t Indicates action a in the current state t The obtained photosynthetic response feedback; action a t Indicates that in state z t The action to be performed is selected below; z t+1 This represents the state trajectory vector of the system at the next moment after the action is performed; γ∈[0,1] is the discount factor; Represents the complete set of actions; a′ represents the candidate action for evaluating the maximum Q value in the next state; Q(z) t ,a t ) indicates that in state z t Next, execute action a t The expected regulatory value that can be obtained; Indicates that in state z t+1 The action with the highest expected value among all candidate actions is evaluated; T represents the total number of time steps used for training; y t Indicates based on the current feedback r t The target control value is calculated based on the maximum action value of the future state. Using optimizers based on loss functions Backpropagation and gradient updates are performed on the parameters θ of the scoring network, enabling the recursive network to gradually optimize the hormone regulation strategy in the state-action space.

8. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 7, characterized in that: In S4, the execution time t of each action in the state trajectory is determined. a The actual peak response time t in the corresponding photosynthetic response data r Perform time difference calculations to obtain the actual response delay value Δt = t for each action in the state trajectory. r -t a ; The actual response delay value is matched with the control prediction window corresponding to the action trigger time to determine whether the current control behavior exceeds the policy prediction range, and the delay deviation label matrix is ​​output. Based on the delay bias label matrix, in the original feedback value r t Apply a time-sensitive function adjustment to construct a delay-sensitive feedback value. Where γ∈(0,1) is the delay discount factor; By z t a t and Combining to construct new training triples Training Triples Objectives used to regulate the value function; Using the training triples Perform target regression calculation of the state-action mapping function in the deep Q network to update the expected regulatory value estimate of each action under different states; During training, the distribution of delayed behavior at consecutive time steps is recorded, and a dynamic response window scheduling strategy is constructed based on the cumulative trend of delay bias to correct the time tolerance interval for future action selection in the Q network.

9. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 8, characterized in that: In S5, by inputting the current state of the rootstock into the trained delayed-aware Q-network model, the action regulation value distribution corresponding to the current state is extracted. The state includes the rootstock hormone level, environmental parameters, and residual effect of the previous action. The extracted action regulation value distribution is dynamically mapped to a preset value evaluation function, wherein the value evaluation function is dynamically updated based on the historical state-action-reward sequence, and the regulation value of each action is normalized during the mapping process. All hormone-releasing actions in the action set are sorted according to the normalized regulatory value, and the effect stability confidence of each hormone-releasing action in the current state is calculated. The confidence is obtained based on the response delay variance corresponding to similar states in the training phase. From the ranking results, select the hormone release actions with the greatest regulatory value and confidence level above the preset threshold, and extract their corresponding hormone type, release time and duration of action to construct a subset of state-associated hormone regulatory actions.

10. The method for regulating hormone signal transduction between grafting rootstock and scion to enhance photosynthetic efficiency according to claim 9, characterized in that: In S5, each action in the candidate subset of hormone regulation actions is combined with current hormone reserve data and recent environmental prediction data into an input data package; The input data packet is fed into the existing strategy generation module, where it is determined whether each hormone action meets the current hormone reserve requirements and whether it is suitable for the predicted environmental conditions. Select the action plan that meets the requirements as the current control plan, and finally output a set of target control plans that meet the requirements of resource constraints and environmental adaptability; The hormone release control device, according to the hormone release type, release time and release dose specified in the regulation scheme, completes the quantitative hormone release operation on the rootstock, thereby achieving target-oriented regulation of the photosynthetic behavior of the scion.

Images