A method for regulating bud differentiation of phalaenopsis by using day-night temperature difference

Abstract

The present application provides a kind of regulation method for inducing phalaenopsis flower bud differentiation by day and night temperature difference, belong to plant culture technical field, the present application is by collecting each strain phalaenopsis genome single nucleotide polymorphism marker and historical phenotype data constructs temperature threshold prediction model, formulates differential day and night temperature control scheme and embeds optimal cooling curve planning module;During induction, use stem flow sensor and leaf microclimate probe to collect stem flow rate and leaf water potential in real time;Multi-channel physiological signal is input into day and night rhythm perception temperature control network, and the flower probability prediction value and target control temperature parameter are output in real time by artificial intelligence;With the flower rate and refrigeration energy consumption as double target, the energy-saving flowering game optimization model is constructed, the pareto frontier is solved on the agent surface, and the result is written back to the temperature control scheme;The technical problem that temperature control decision cannot simultaneously perceive plant rhythm physiological state and real-time water stress state, leading to the disconnection of temperature control strategy and actual physiological demand of plant is solved.

Description

Technical Field

[0001] This invention belongs to the field of plant culture technology, and more specifically, relates to a method for regulating the differentiation of Phalaenopsis orchid buds by utilizing diurnal temperature variation. Background Technology

[0002] Phalaenopsis flower bud differentiation exhibits a strict physiological dependence on diurnal temperature variation. Traditional low-temperature-induced flowering techniques rely on manually setting fixed nighttime temperature ranges and cooling durations, using empirical parameters to drive refrigeration equipment to execute cooling operations according to preset programs. Existing temperature control systems are widely used in mass flowering production in greenhouses, typically employing temperature and humidity sensors to achieve basic closed-loop feedback control. Some systems incorporate historical data from different Phalaenopsis varieties for categorized management of nighttime temperature limits. However, due to genetic differences in temperature-sensing responses among different Phalaenopsis varieties, fixed-parameter schemes are difficult to adapt to parallel production scenarios involving multiple varieties. Furthermore, existing control methods only collect environmental temperature and humidity signals, failing to detect internal physiological states such as stem flow rate and leaf water potential. During continuous low-temperature treatment, when the temperature drop exceeds the plant's water conduction capacity regulation limit, a simultaneous decrease in stem flow rate and leaf water potential occurs, leading to water stress and necrosis of the original flower bud genes. However, existing temperature control systems lack both a coupled stress perception mechanism and dynamic decision-making capabilities based on the plant's rhythmic physiological state, making it impossible to automatically switch between normal temperature control and emergency protection. In other words, existing technologies have a technical problem: during the diurnal temperature variation-induced flower bud differentiation process of Phalaenopsis orchids, temperature control decisions cannot simultaneously perceive the plant's rhythmic physiological state and real-time water stress state, leading to a disconnect between temperature control strategies and the plant's actual physiological needs. Summary of the Invention

[0003] In view of this, the present invention provides a method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation, which can solve the problem of...

[0004] Existing technologies have a technical problem: during the diurnal temperature variation-induced flower bud differentiation process of Phalaenopsis orchids, temperature control decisions cannot simultaneously perceive the plant's rhythmic physiological state and real-time water stress state, leading to a disconnect between temperature control strategies and the plant's actual physiological needs.

[0005] This invention is implemented as follows: This invention provides a method for regulating the differentiation of Phalaenopsis orchid buds by utilizing diurnal temperature variation, comprising the following steps:

[0006] We collected genomic single nucleotide polymorphism markers and historical phenotypic data of Phalaenopsis orchids from various strains, constructed a strain temperature sensitivity threshold prediction model, and generated a temperature sensitivity threshold label for each strain.

[0007] Based on the temperature threshold label, a differentiated day and night temperature control scheme is formulated, and an optimal cooling curve planning module is embedded to control the cooling rate within the cooling rate range.

[0008] During the diurnal temperature range induction period, stem flow sensor and leaf microclimate probe are used to collect stem flow rate and leaf water potential in real time, and a temperature-water potential coupled dynamic model is established. When the coupled stress index exceeds the coupled stress triggering threshold, the temperature slow rise program and water replenishment control are triggered in conjunction.

[0009] Temperature time series data, stem flow rate, leaf water potential and stomatal conductance are input into the diurnal rhythm sensing temperature control network. The diurnal rhythm sensing temperature control network outputs the predicted value of flowering probability, the target night temperature setpoint and the target cooling rate in real time. The target night temperature setpoint and the target cooling rate are written into the temperature control execution module.

[0010] With flowering rate and cooling energy consumption as dual objectives, an energy-saving flowering game optimization model is constructed. The Pareto front is solved on the surrogate surface, and the night temperature setpoint and cooling rate corresponding to the Pareto front are written back to the differentiated day and night temperature control scheme.

[0011] Based on the temperature threshold label and real-time collected stem flow rate, leaf water potential, stomatal conductance and temperature time series data, the weight parameters of the diurnal rhythm sensing temperature control network are periodically updated to complete closed-loop adaptive regulation.

[0012] Specifically, the acquisition of the temperature-sensing threshold labels involves conducting orthogonal experiments on representative varieties under multiple low-temperature treatment days and multiple night temperature setpoints, recording the flowering rate, determining the temperature-sensing threshold of each variety through regression analysis, forming a variety temperature-sensing threshold database, and using the variety temperature-sensing threshold database as the training label for the variety temperature-sensing threshold prediction model.

[0013] Specifically, the cooling rate range of the optimal cooling curve planning module is determined by collecting transcriptome data from plants at each cooling rate level through gradient cooling experiments, analyzing the response relationship between heat shock protein gene expression and flowering rate, fitting the response surface between cooling rate and flowering rate, and taking the cooling rate range where the flowering rate is higher than the flowering rate threshold and the heat shock protein gene expression is lower than the baseline value by a certain proportion as the safe cooling rate range.

[0014] The coupled stress index is calculated from the stem flow rate and the leaf water potential. Specifically, it is obtained by multiplying the ratio of the real-time stem flow rate to the stem flow rate baseline by the ratio of the real-time leaf water potential to the leaf water potential baseline.

[0015] Specifically, the coupling stress triggering threshold is obtained by recording the necrosis rate of flower bud primordia under different combinations of temperature jump amplitude and different initial water deficit states. The coupling stress index value corresponding to the flower bud primordia necrosis rate exceeding the necrosis rate threshold is used as the quantile of the coupling stress triggering threshold after multiple batches of repeated experiments.

[0016] The circadian rhythm sensing temperature control network includes a pulse-coded rhythmic stream, a dynamic graph continuous stream, and a cross-modal attention fusion layer. Temperature time-series data is converted into discrete pulse sequences according to a rate of change threshold and then input into the pulse-coded rhythmic stream. Stem flow rate, leaf water potential, stomatal conductance, and steady-state temperature value are used as dynamic graph node features and input into the dynamic graph continuous stream. The two outputs are fused by the cross-modal attention fusion layer and then input into the differentiation prediction head and the temperature control strategy head, respectively.

[0017] The pulse-coded rhythm stream is composed of a stack of multiple leaky integral firing neurons. The interlayer synaptic plasticity rules are used for unsupervised pre-training, and the interlayer synaptic weight matrix is ​​sparsely initialized.

[0018] The circadian rhythm sensing temperature control network introduces a rhythm adjustment function, which takes the coupling stress index, the predicted flowering probability, and the normalized value of the effective cumulative number of low temperature days as inputs to calculate the rhythm adjustment value. Based on the rhythm adjustment value, the leakage coefficient of the leaking integral firing neurons in the pulse-coded rhythm stream is dynamically adjusted.

[0019] The circadian rhythm sensing temperature control network is equipped with a jump mechanism. When an abnormal circadian rhythm pulse pattern is detected, the normal output of the temperature control strategy head is skipped, and the emergency protection control sub-network is directly activated to output the temperature rise rate and water replenishment amount.

[0020] The dynamic graph continuous flow is composed of a multi-layer graph convolutional network. The adjacency matrix of the dynamic graph is dynamically updated according to the plant growth and development stage, which is determined by a low-temperature treatment day counter.

[0021] The energy-saving flowering game optimization model is a two-layer game structure. The upper-layer model aims to maximize the flowering rate, while the lower-layer model aims to minimize the cooling energy consumption. The two objective functions are related through coupling terms.

[0022] The surrogate surface is obtained by fitting a Gaussian process regression surrogate model with a small number of real simulation samples, and the Pareto front is solved by using a reference point-guided non-dominated sorting genetic algorithm to quickly approximate the surrogate surface.

[0023] Specifically, the upper-level coupling weight coefficient in the upper-level objective function and the lower-level coupling weight coefficient in the lower-level objective function are combined under different varieties and different production batches, and the optimal combination is selected and confirmed based on the criterion of maximizing the hypervolume index of the Pareto front on the validation set.

[0024] Specifically, the training of the circadian rhythm sensing temperature control network involves first unsupervised pre-training of the pulse-coded rhythm stream using time-dependent synaptic plasticity rules, then supervised pre-training of the dynamic graph continuous stream using cross-entropy loss, and finally joint fine-tuning of the entire circadian rhythm sensing temperature control network. The loss function is a weighted sum of differentiation prediction loss and temperature control strategy loss.

[0025] Among them, the night temperature setpoint range of the differentiated day and night temperature control scheme is 18-22℃, the number of days of low temperature treatment ranges from 14 to 21 days, and the cooling rate ranges from 0.5 to 3℃ / h; in the leakage coefficient adjustment range of the rhythm regulation value, the leakage coefficient ranges from 0.80 to 0.85 when the rhythm regulation value is not lower than 0.8, the leakage coefficient ranges from 0.85 to 0.90 when the rhythm regulation value is between 0.5 and 0.8, and the leakage coefficient ranges from 0.90 to 0.95 when the rhythm regulation value is lower than 0.5; the coupled stress trigger threshold is taken as the 95th quantile, the necrosis rate threshold is 10%, the joint fine-tuning weighting coefficient ranges from 0.3 to 0.7, and the initial value of the learning rate ranges from 0.0005 to 0.002.

[0026] This invention constructs a diurnal rhythm-sensing temperature control network, structurally modeling temperature pulse events and plant physiological regulation networks using pulse-coded rhythmic flow and dynamic graph continuous flow, respectively. These two types of features are then integrated through a cross-modal attention fusion layer, enabling temperature control decisions to simultaneously possess time-awareness and physiological structure-awareness. When the plant's water stress triggers a coupled stress index exceeding a threshold, the diurnal rhythm-sensing temperature control network's switching mechanism automatically activates the emergency protection control sub-network, outputting the temperature rise rate and water replenishment amount. This protects flower bud primordia from water stress damage without interrupting the low-temperature induction process. The energy-saving flowering game optimization model further solves the Pareto front between flowering rate and cooling energy consumption, writing the optimal nighttime temperature setpoint and cooling rate back to the differentiated diurnal temperature control scheme, achieving continuous adaptive optimization of temperature control parameters. In summary, this invention solves the technical problem mentioned in the background art: during the diurnal temperature difference-induced flower bud differentiation process in Phalaenopsis orchids, temperature control decisions cannot simultaneously perceive the plant's rhythmic physiological state and real-time water stress state, leading to a disconnect between temperature control strategies and the plant's actual physiological needs. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention.

[0028] Figure 2 This is a graph showing the coordinated changes in the leakage coefficient adjustment trajectory and the predicted flowering probability after the emergency protection of strain C is triggered.

[0029] Figure 3 Optimize the Pareto front distribution map for energy-saving flowering game of each strain.

[0030] Figure 4The convergence plot of the mean square error for predicting flowering probability in the diurnal rhythm sensing temperature control network after updating the weights for each period. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0032] like Figure 1 The diagram shown is a flowchart of a method for regulating Phalaenopsis orchid bud differentiation using diurnal temperature variation, provided by this invention. This method includes the following steps:

[0033] S01. Collect single nucleotide polymorphism markers and historical phenotypic data of the genome of each strain of Phalaenopsis orchid, construct a strain temperature sensitivity threshold prediction model, and generate a temperature sensitivity threshold label for each strain;

[0034] S02. Based on the temperature threshold label, formulate a differentiated day and night temperature control scheme, set the night temperature setpoint in the range of 18 to 22℃, set the number of days of low temperature treatment in the range of 14 to 21 days, and embed the optimal cooling curve planning module to control the cooling rate in the range of 0.5 to 3℃ / h.

[0035] S03. During the diurnal temperature difference induction period, the stem flow sensor and leaf microclimate probe are used to collect the stem flow rate and leaf water potential in real time, and a temperature-water potential coupled dynamic model is established. When the coupled stress index exceeds the coupled stress triggering threshold, the temperature slow rise program and water replenishment control are triggered in conjunction.

[0036] S04. Input the temperature time series data, stem flow rate, leaf water potential and stomatal conductance into the diurnal rhythm sensing temperature control network. The diurnal rhythm sensing temperature control network outputs the predicted value of flowering probability, the target night temperature setpoint and the target cooling rate in real time. Then write the target night temperature setpoint and the target cooling rate into the temperature control execution module.

[0037] S05. With flowering rate and cooling energy consumption as dual objectives, construct an energy-saving flowering game optimization model, quickly solve the Pareto front on the proxy surface, and write the night temperature setpoint and cooling rate corresponding to the Pareto front back to the differentiated day and night temperature control scheme.

[0038] S06. Based on the temperature threshold label and the real-time collected stem flow rate, leaf water potential, stomatal conductance and temperature time series data, the weight parameters of the diurnal rhythm sensing temperature control network are periodically updated to complete closed-loop adaptive regulation.

[0039] The single nucleotide polymorphism marker is a genetic variation site at a single base position in the genome, used to describe the genetic differences in temperature-sensing responses among different strains of Phalaenopsis orchids.

[0040] The temperature-sensing threshold label is the minimum effective number of days of low-temperature treatment and the lower limit of night temperature required for each cultivar, predicted based on single nucleotide polymorphism markers and historical phenotypic data. The temperature-sensing threshold label is obtained by conducting orthogonal experiments on 30-50 representative cultivars under low-temperature treatment conditions of 14, 17, and 21 days and night temperature settings of 18℃, 20℃, and 22℃, recording the flowering rate, determining the temperature-sensing threshold for each cultivar through regression analysis, forming a cultivar temperature-sensing threshold database, and using this database as the training label for the cultivar temperature-sensing threshold prediction model.

[0041] The optimal cooling curve planning module is a parameterized cooling path planning program embedded in the temperature control execution module. The cooling rate range of 0.5–3℃ / h is derived as follows: a 5×6 gradient cooling experiment matrix was designed, with cooling rates at five levels: 0.5℃ / h, 1℃ / h, 1.5℃ / h, 2℃ / h, and 3℃ / h. Transcriptome data were collected from plants at each level, and the response relationship between heat shock protein gene expression and flowering rate was analyzed. A response surface between cooling rate and flowering rate was fitted, and the cooling rate range where the flowering rate was higher than 85% and the heat shock protein gene expression was lower than the baseline value by 20% was selected as the safe cooling rate range. The experiment was repeated three times, and the average value was used to confirm the boundary value.

[0042] The temperature-water potential coupled dynamic model is a set of simultaneous dynamic equations describing the changes in stem flow rate and leaf water potential with temperature. The coupled stress index is calculated from the stem flow rate and leaf water potential, and the formula is as follows:

[0043] ;

[0044] in The coupling stress index is dimensionless. The real-time stem flow rate is (g / h). The stem flow rate is the baseline value (g / h) under normal growth conditions. Real-time blade water potential (MPa). The leaf water potential (MPa) is the baseline value under normal growth conditions. The coupled stress triggering threshold is obtained by recording the flower bud primordium necrosis rate under different combinations of temperature jumps (4–10℃) and initial water deficit states, with the threshold value corresponding to a flower bud primordium necrosis rate exceeding 10%. The value was used as the coupling stress triggering threshold, and the 95th percentile was taken as the final coupling stress triggering threshold after 5 batches of repeated experiments.

[0045] The circadian rhythm sensing temperature control network is a prediction and decision-making model based on deep learning, and its structure is divided into three parts as described below.

[0046] The first segment consists of the input layer and the pulse-coded rhythm stream. The input to the diurnal rhythm sensing temperature control network consists of four signals: temperature time-series data from the temperature control execution module, stem flow rate from the stem flow sensor, leaf water potential from the leaf microclimate probe, and stomatal conductance from the leaf microclimate probe. The temperature time-series data is converted into a discrete pulse sequence according to a rate of change threshold. The rate of change threshold is obtained by statistically analyzing the temperature change rate distribution of normal cooling events and abnormal stress cooling events in historical temperature records, using the intersection of the two distributions as the rate of change threshold, and confirming it through three batches of experimental iterations. The discrete pulse sequence input is a spiking neural network consisting of three layers of leaky integral firing neurons stacked together. The first layer has 128 leaky integral firing neurons, the second layer has 64, and the third layer has 32. The initial value of the leakage coefficient is set in the range of 0.8 to 0.95. The inter-layer synaptic weight matrix is ​​sparsely initialized with a sparsity of 80% to 90%. Sparse initialization reduces the memory usage of each layer of leaky integral firing neurons to 10% to 20% of that of dense initialization. The pulse-coded rhythm stream occupies a dedicated CUDA stream to ensure the temporal consistency of event-driven inference. Inter-layer unsupervised pre-training is performed using time-dependent synaptic plasticity rules. The pulse-coded rhythm stream outputs a rhythm pulse feature vector with a dimension of 32.

[0047] The second section is a dynamic graph continuous flow and cross-modal attention fusion layer. Four physiological indicators—stem flow rate, leaf water potential, stomatal conductance, and steady-state temperature values ​​from time-series temperature data—are used as dynamic graph node features. Known physiological regulatory relationships are used as dynamic graph edges. The adjacency matrix of the dynamic graph is dynamically updated according to the plant's growth and development stage, which is determined by a counter for the number of days of low-temperature treatment. The dynamic graph continuous flow consists of a three-layer graph convolutional network: the first layer has a hidden layer dimension of 64, the second layer has 32, and the third layer has 16. The adjacency matrix of the graph convolutional network is stored in blocks to adapt to parallel inference across multiple varieties. CUDA thread blocks are dynamically divided in batches based on the dynamic graph node size. The dynamic graph continuous flow outputs a physiological state feature vector with a dimension of 16. The rhythmic pulse feature vector and the physiological state feature vector are input into the cross-modal attention fusion layer. The cross-modal attention fusion layer has 4 attention heads and outputs a fused feature vector with a dimension of 64. The video memory of the cross-modal attention fusion layer is allocated by the pulse-coded rhythm stream and the dynamic graph continuous stream sharing a memory pool. The capacity of the shared memory pool is set to 1.5 times the peak video memory of a single branch. The data loop adopts a double-buffered queue, and forward inference and data prefetching are executed in parallel. The depth of the double-buffered queue is set in the range of 8 to 16 frames.

[0048] The third section comprises the output head and the jump mechanism. The fused feature vector is simultaneously input to both the differentiation prediction head and the temperature control strategy head. The differentiation prediction head consists of a two-layer fully connected network, outputting a predicted flowering probability value (dimensionless, ranging from 0 to 1). The temperature control strategy head also consists of a two-layer fully connected network, outputting a target nighttime temperature setpoint (°C) and a target cooling rate (°C / h). The target nighttime temperature setpoint is written to the temperature control execution module, and the target cooling rate is written to the optimal cooling curve planning module. The jump mechanism determines the presence of abnormal rhythmic pulse patterns based on the rhythmic pulse feature vector output by the pulse-coded rhythmic flow: when an abnormal rhythmic pulse pattern is detected, the normal output of the temperature control strategy head is skipped, and the emergency protection control sub-network is directly activated. This emergency protection control sub-network is a single-layer fully connected network that takes the fused feature vector as input and outputs a temperature rise rate (°C / h) and a water replenishment volume (mL). The temperature rise rate is written to the temperature control execution module, and the water replenishment volume is written to the water replenishment control module.

[0049] The circadian rhythm sensing temperature control network introduces a rhythm adjustment function to dynamically adjust the leakage coefficient of the leaking integral firing neurons in the pulse-coded rhythm stream. The rhythm adjustment function takes the coupling stress index, the predicted flowering probability, and the normalized value of the effective cumulative number of low-temperature days as inputs to calculate the rhythm adjustment value, as expressed in the following formula:

[0050] ;

[0051] in This is the rhythm regulation value (dimensionless). The coupling stress index is dimensionless. The coupling stress trigger threshold (dimensionless). This is a predicted value for the probability of flowering (dimensionless, ranging from 0 to 1). This represents the current cumulative number of days with effective low temperatures. The number of days (in days) for the target low-temperature treatment of the strain in the temperature threshold label. , , The weighting coefficients and ;when At this time, the leakage coefficient is lowered to the range of 0.80 to 0.85 to enhance the sensitivity of the pulse-coded rhythmic flow to temperature pulse events; when At that time, maintain the leakage coefficient within the range of 0.85 to 0.90 to maintain a normal rhythmic sensing state; when At this time, the leakage coefficient is increased to the range of 0.90 to 0.95 to reduce the sensitivity of the pulse-coded rhythm flow to brief temperature disturbances and avoid frequent false triggering of the jump mechanism. The weighting coefficient... , , The acquisition method is as follows: on the training set, the Pearson correlation coefficient between the rhythm regulation value and the flower bud differentiation result is maximized as the criterion, and the data is traversed in the range of 0 to 1 with a step size of 0.1. , , The optimal combination was determined by repeating the experiment three times and taking the average value.

[0052] The steps for establishing the training dataset of the circadian rhythm sensing temperature control network specifically include: collecting time-series temperature data, stem flow rate records, leaf water potential records, stomatal conductance records, and corresponding flowering rate records for 30-50 varieties of Phalaenopsis orchids under different night temperature setpoints, cooling rates, stem flow rates, and leaf water potential combinations; dividing the dataset into training, validation, and test sets in an 8:1:1 ratio; generating discrete pulse sequence annotations for the time-series temperature data based on the rate of change threshold; constructing dynamic graph topology annotations for stem flow rate, leaf water potential, stomatal conductance, and steady-state temperature values ​​according to the plant's growth and development stages; and manually labeling jump trigger tags for abnormal stress events.

[0053] The specific steps for training the circadian rhythm sensing temperature control network include: first, unsupervised pre-training of the pulse-coded rhythmic flow using time-dependent synaptic plasticity rules, with 100-200 training rounds; then, supervised learning pre-training of the dynamic graph continuous flow using cross-entropy loss function, with 50-100 training rounds; finally, joint fine-tuning of the entire circadian rhythm sensing temperature control network, with the loss function being the weighted sum of differentiation prediction loss and temperature control strategy loss, the weighting coefficients being determined by searching according to the optimal validation ensemble flower rate principle within the range of 0.3-0.7, with 100-300 training rounds, and the optimizer using the adaptive moment estimation algorithm, with the initial learning rate set within the range of 0.0005-0.002.

[0054] The circadian rhythm sensing temperature control network maintains high sensitivity to discrete temperature pulse events through event-driven sparse computation of pulse-coded rhythmic flow. It performs structured modeling of the stage-wise topological changes of the plant's physiological regulation network through dynamic graph continuous flow, enabling temperature control decisions to have both time-aware and physiological structure-aware capabilities. These two capabilities complement each other through a cross-modal attention fusion layer, automatically switching between normal regulation and emergency protection control, thereby improving the accuracy of flowering probability prediction and reducing the false control rate.

[0055] The energy-saving flowering game optimization model is a two-layer game structure. The upper-layer model aims to maximize the flowering rate, while the lower-layer model aims to minimize cooling energy consumption. The two objective functions are related through coupling terms. The upper-layer objective function is expressed as follows:

[0056] ;

[0057] in This is the target value for the upper layer (dimensionless). To predict the flowering rate (%) This represents the highest flowering rate (%) in the history of this variety. Set the nighttime temperature (°C). The cooling rate is expressed as °C / h. The energy consumption is for cooling (kW·h). The baseline cooling energy consumption is (kW·h). The upper-layer coupling weight coefficients are dimensionless, and the constraints are as follows: ℃, ℃ / h. The lower-level objective function is expressed as follows:

[0058] ;

[0059] in The target value for the lower level is dimensionless. The lower-level coupling weight coefficients are dimensionless, and the constraints are as follows: , The minimum flowering rate (%) required for production. Both objective functions have coupling terms containing... and This creates a game-like relationship between the upper and lower targets regarding the nighttime temperature setpoint and the cooling rate. A Gaussian process regression surrogate model is used to replace real simulation evaluation, and a non-dominated sorting genetic algorithm guided by reference points is combined on the surrogate surface to quickly approximate the Pareto front. The upper-layer coupling weight coefficient... Coupling weight coefficients with lower layers The acquisition method is as follows: for different strains and different production batches, traverse within the range of 0.1 to 0.5 with a step size of 5%. and The optimal combination was selected based on maximizing the hypervolume index at the Pareto front on the validation set, and the results were repeated in three batches to confirm the optimal combination.

[0060] The upper-level objective function is used to maximize the flowering rate while ensuring that the cooling energy consumption penalty is acceptable. Its inputs include the nighttime temperature setpoint and the cooling rate, and its output is the upper-level objective value. The lower-level objective function is used to minimize the cooling energy consumption while ensuring that the flowering rate is not lower than the minimum flowering rate constraint. Its inputs include the nighttime temperature setpoint and the cooling rate, and its output is the lower-level objective value.

[0061] The effective cumulative low-temperature days refer to the cumulative number of days since the diurnal temperature difference induction began, during which the nighttime temperature setpoint has consistently been lower than the lower limit of the nighttime temperature in the strain's temperature-sensing threshold label. The surrogate surface is an approximate surface of the objective function obtained by fitting a Gaussian process regression surrogate model with a small number of real simulation samples. It is used to replace real simulation evaluation to significantly reduce the computational cost of solving the Pareto front. The reference-point-guided non-dominated sorting genetic algorithm is a multi-objective evolutionary algorithm that guides the population to uniformly distribute towards the Pareto front by pre-setting reference points in the objective space. It is used to solve the multi-objective optimization problem of nighttime temperature setpoint and cooling rate on the surrogate surface. The leakage integral firing neuron is the basic computational unit in the pulse-coded rhythm stream. Its membrane potential decays exponentially with time according to the leakage coefficient. When the membrane potential exceeds the discharge threshold, it generates a pulse and resets, used for time-encoding discrete temperature pulse events. The time-dependent synaptic plasticity rule is an unsupervised learning rule that adjusts synaptic weights based on the pulse firing time difference between the leakage integral firing neurons before and after the synapse. It is used to pre-train the pulse-coded rhythm stream on unlabeled data to extract diurnal rhythm pulse patterns. The dynamic graph topology is a graph structure with plant physiological indicators as nodes, known physiological regulatory relationships as edges, and an adjacency matrix that is updated with the plant's growth and development stages. It is used to capture the phased changes in the plant's physiological state. The cross-modal attention fusion layer is a network layer that calculates attention weights for the rhythmic pulse feature vector and the physiological state feature vector separately, and then weights and fuses them. It is used to coordinate the integration of information between pulse temporal features and graph structure features. The abnormal rhythmic pulse pattern is the pulse distribution state in the rhythmic pulse feature vector output by the pulse-encoded rhythm stream, where the pulse firing frequency per unit time exceeds the normal diurnal rhythm range. It is used to identify the plant's stress response triggering jump mechanism. The emergency protection control sub-network is a single-layer fully connected network activated within the diurnal rhythm sensing temperature control network when an abnormal rhythmic pulse pattern is detected. It takes the fused feature vector as input and outputs the temperature rise rate and water replenishment amount. It is used to replace the normal output of the temperature control strategy head under plant stress. The dual-buffered queue is a two-stage alternating buffer structure in which forward inference and data prefetching are executed in parallel during the data loop. It is used to reduce the inference waiting latency of the diurnal rhythm sensing temperature control network. The flower bud primordium necrosis rate is the proportion of necrotic flower bud primordia per unit area, used to characterize the degree of damage to flower bud differentiation quality caused by dual stress in the temperature-water potential coupled dynamic model.

[0062] The specific implementation of step S01 is as follows: First, genomic DNA is extracted from 30-50 representative Phalaenopsis orchid strains. High-throughput sequencing technology is used to obtain whole-genome single nucleotide polymorphism (SNP) marker site data, forming a high-dimensional genetic feature vector for each strain. Simultaneously, historical flowering rate records for each strain under different low-temperature treatment conditions are collected to form a phenotypic dataset. Using the genetic feature vectors of each strain as input and the historical flowering rate as a supervision label, a regularized regression model is used to fit the mapping relationship between genotype and temperature-sensing response. This generates a temperature-sensing threshold label for each strain, containing the minimum effective number of days of low-temperature treatment and the lower limit of nighttime temperature. The temperature-sensing thresholds of all strains are then compiled into a strain temperature-sensing threshold database, serving as training labels for the strain temperature-sensing threshold prediction model. The temperature-sensing threshold labels directly determine the parameter boundaries of subsequent differentiated temperature control schemes and are a prerequisite for multi-strain parallel control.

[0063] The specific implementation of step S02 is as follows: Based on the temperature threshold tag generated in step S01, a nighttime temperature setpoint and a low-temperature treatment period are set for each strain. The nighttime temperature setpoint ranges from 18 to 22°C, and the low-temperature treatment period ranges from 14 to 21 days, forming a differentiated day-night temperature control scheme. The optimal cooling curve planning module is embedded in the temperature control execution module, and its safe cooling rate range is determined through a 5×6 gradient cooling experiment matrix: the cooling rate is set to 0.5... 1 1.5 2 3 Five levels were used in the study. Transcriptome data of plants were collected at each level to analyze the response relationship between heat shock protein gene expression and flowering rate. After fitting the response surface, the cooling rate range in which the flowering rate was higher than 85% and the heat shock protein gene expression was lower than the baseline value by 20% was selected as the safe cooling rate range. The experiment was repeated three times and the mean value was used to confirm the boundary value. Finally, the cooling rate was controlled between 0.5 and 3. Within the specified range, the optimal cooling curve planning module generates a parameterized cooling path based on the above boundary values, ensuring that the plant is not subjected to heat shock damage from rapid cooling throughout the entire low-temperature induction period.

[0064] The specific implementation of step S03 is as follows: During the diurnal temperature variation induction period, the stem flow sensor continuously collects stem flow rate data, and the leaf microclimate probe continuously collects leaf water potential and local temperature data. The temperature-water potential coupled dynamic model uses a system of simultaneous dynamic equations to describe the coordinated changes in stem flow rate and leaf water potential with temperature. Coupled stress index. From real-time stem flow rate Compared with stem flow rate benchmark value The ratio multiplied by the real-time blade water potential Compared with the blade water potential baseline value The ratio was calculated. The coupled stress triggering threshold was determined by recording the primordium necrosis rate under different combinations of temperature jumps (4–10 °C) and different initial water deficit states. The threshold was defined as the value corresponding to a primordium necrosis rate exceeding 10%. The value was repeated in 5 batches, and the 95th percentile was used as the final threshold. When the coupled stress trigger threshold is exceeded, the system triggers a temperature slow-rise program and water replenishment control to prevent the flower buds from dying due to dual stress.

[0065] The specific implementation of step S04 is as follows: The diurnal rhythm sensing temperature control network receives four input signals, namely temperature time-series data, stem flow rate, leaf water potential, and stomatal conductance. The temperature time-series data is converted into a discrete pulse sequence according to the rate of change threshold and then input into the pulse-coded rhythm stream. The pulse-coded rhythm stream is composed of three layers of leaky integral firing neurons stacked together. The interlayer connections are pre-trained unsupervised using time-dependent synaptic plasticity rules, and the interlayer synaptic weight matrix is ​​sparsely initialized with a sparsity of 80% to 90%. The stem flow rate, leaf water potential, stomatal conductance, and steady-state temperature value are used as dynamic graph node features input into the dynamic graph continuous stream. The dynamic graph continuous stream is composed of three layers of graph convolutional network, and the adjacency matrix is ​​updated with the low-temperature treatment day counter. The pulse-coded rhythmic stream outputs a rhythmic pulse feature vector, while the dynamic graph continuous stream outputs a physiological state feature vector. These two feature vectors are fused through a cross-modal attention fusion layer (with four attention heads) and then input into the differentiation prediction head (outputting a predicted flowering probability) and the temperature control strategy head (outputting a target nighttime temperature setpoint and target cooling rate), respectively, and written into the temperature control execution module. The rhythm regulation function takes the coupling stress index, the predicted flowering probability, and the normalized value of the effective accumulated low-temperature days as inputs, dynamically adjusting the leakage coefficient of the leakage integral firing neuron to make the sensitivity of the pulse-coded rhythmic stream adaptively change with the plant's physiological state. When an abnormal rhythmic pulse pattern is detected, a jump mechanism activates the emergency protection control subnetwork, outputting the temperature rise rate and water replenishment amount.

[0066] The specific implementation of step S05 is as follows: The energy-saving flowering game optimization model adopts a two-layer game structure. The upper-layer objective function maximizes the flowering rate under the premise that the cooling energy consumption penalty is acceptable, while the lower-layer objective function minimizes the cooling energy consumption under the constraint that the flowering rate is not lower than the minimum flowering rate. Both objective functions use the nighttime temperature setpoint and the cooling rate as decision variables, and form a mutually restrictive game relationship through a coupling term that includes the ratio of predicted flowering rate to cooling energy consumption. During the solution process, a Gaussian process regression surrogate model is used to fit the approximate surface of the objective function with a small number of real simulation samples. Then, a non-dominated sorting genetic algorithm guided by reference points is combined on the surrogate surface to quickly approximate the Pareto front, significantly reducing the computational cost. The nighttime temperature setpoint and cooling rate corresponding to the Pareto front are written back to the differentiated day-night temperature control scheme to achieve continuous optimization of the temperature control parameters. Upper-layer coupling weight coefficients... Coupling weight coefficients with lower layers The maximum excess volume index was used as the criterion for confirmation through traversal experiments under different strains and production batches.

[0067] The specific implementation of step S06 is as follows: using the temperature threshold label as the strain-level monitoring signal, and combining real-time collected stem flow rate, leaf water potential, stomatal conductance, and temperature time-series data, the weight parameters of the diurnal rhythm sensing temperature control network are periodically updated. In the update strategy, the pulse-coded rhythmic flow is updated locally without supervision using time-dependent synaptic plasticity rules, and the dynamic graph continuous flow and the joint fine-tuning of the entire network use an adaptive moment estimation algorithm. The loss function is the weighted sum of differentiation prediction loss and temperature control strategy loss, and the weighting coefficients are searched and determined within the range of 0.3 to 0.7 according to the principle of optimal flower rate verification integration. The initial learning rate is set within the range of 0.0005 to 0.002. Periodic weight updates enable the diurnal rhythm sensing temperature control network to continuously adapt to different strains, batches, and environmental conditions, forming a complete closed-loop adaptive control mechanism.

[0068] It should be noted that the key technologies of this invention include: First, a heterogeneous dual-flow modeling mechanism of pulse-coded rhythmic flow and dynamic graph continuous flow in the diurnal rhythm sensing temperature control network. The pulse-coded rhythmic flow captures the temporal rhythmic characteristics of temperature pulse events through event-driven sparse computation, while the dynamic graph continuous flow captures the structural changes of the plant's physiological regulation network through staged graph topology updates. Neither of them can independently achieve joint sensing of the plant's rhythmic physiological state and water stress state. Second, an adaptive leakage coefficient adjustment mechanism driven by the rhythm regulation function. This mechanism uses the coupled stress index, the predicted value of flowering probability, and the effective cumulative number of low temperature days as comprehensive inputs to dynamically adjust the sensitivity of the pulse-coded rhythmic flow, avoiding false triggering or missed triggering under stress conditions with fixed sensitivity. Third, a joint solution mechanism of energy-saving flowering game optimization model and Gaussian process regression surrogate surface. By replacing the computationally expensive real simulation evaluation with the surrogate surface, the real-time solution of the Pareto front is feasible in engineering within the greenhouse control cycle. The synergistic effect of these three key technologies elevates temperature control decision-making from a single environmental feedback mechanism to a closed-loop control system that couples rhythmic timing perception, physiological structure perception, and multi-objective parameter optimization at three levels, fundamentally compensating for the lack of physiological perception dimension in existing technologies.

[0069] It should be noted that in the parallel induction of flowering in multiple Phalaenopsis orchid varieties, the low-temperature induction process of different varieties is often at different stages of effective low-temperature accumulation days. Plants simultaneously experience varying degrees of water stress within the same greenhouse, requiring the temperature control system to output differentiated temperature control commands to different varieties at the same time. Furthermore, it needs to independently determine whether to trigger emergency protection for each variety. The reason for these technical problems is that traditional temperature control systems rely on fixed rules or single regression models for their decision units, with input features limited to environmental temperature and humidity. This makes it impossible to distinguish the differentiated physiological responses of different varieties under the same environment, nor can it process the graph structure features and temporal impulse features of multiple varieties in parallel within the same inference frame. This leads to cross-varietal interference in temperature control commands during multi-varietal parallel scenarios and makes it impossible to dynamically adjust the decision boundaries at different growth stages. The usual solution to these technical problems is to deploy an independent rule base or lookup table system for each variety, executing temperature control operations based on preset variety parameters. However, the parameters of the independent rule base method are all static preset values, which cannot respond to the dynamic changes in real-time physiological state. Furthermore, the maintenance cost of the rule base increases linearly with the number of strains. Resource competition between strains during parallel inference also leads to timing delays in temperature control commands, making it impossible to guarantee the real-time performance of commands within the critical time window when the plant is under stress. This invention effectively solves this technical problem. The adjacency matrix of the dynamic graph continuous flow independently maintains the graph topology structure of each strain using a counter for the number of days of low-temperature treatment. The adjacency matrix of the graph convolutional network uses block storage to adapt to parallel inference across multiple strains. The central processing unit thread blocks are dynamically divided in batches of dynamic graph nodes, enabling the extraction of physiological state features from multiple strains to be completed in parallel within the same forward inference frame without cross-strain feature aliasing. The pulse-coded rhythmic flow has an exclusive independent flow in the unified device architecture flow allocation of the graphics processor, ensuring the timing consistency of event-driven inference, thereby ensuring that the temperature control command for each strain is written to the temperature control execution module within the correct time window. The rhythm regulation function adjusts the leakage coefficient corresponding to each cultivar based on the independently calculated coupled stress index and predicted flowering probability. This allows the trigger sensitivity of the switching mechanism to independently and adaptively adapt at the cultivar level, preventing the stress state of one cultivar from mistakenly triggering the emergency protection control subnetwork of other cultivars. This mechanism enables the invention to possess cultivar-level isolation decision-making capabilities in multi-cultivar parallel flowering induction scenarios, fundamentally resolving the contradiction between differentiated decision-making and real-time performance in multi-cultivar parallel temperature control scenarios.

[0070] Specifically, the principle of this invention is:

[0071] The present invention can solve the above-mentioned technical problems. The fundamental reason is that it elevates the plant rhythm physiological perception and real-time water stress perception from the independent signal acquisition level to a unified deep learning decision framework, and realizes closed-loop adaptive control through a multi-level feedback mechanism.

[0072] First, the pulse-coded circadian rhythm stream processes discrete temperature pulse sequences using an event-driven sparse computation approach. The membrane potential of the leaky integral firing neurons decays exponentially over time according to the leakage coefficient, capturing the distribution pattern of temperature change events along the time axis, i.e., the temporal characteristics of the plant's diurnal rhythm response. This mechanism gives the system a natural ability to distinguish between abrupt and gradual temperature changes. Furthermore, the rhythm regulation function dynamically adjusts the leakage coefficient based on the coupling stress index, the predicted flowering probability, and the effective cumulative number of days of low temperature, ensuring that the sensitivity of the pulse-coded circadian rhythm stream adapts to the plant's physiological state and avoids false or missed triggers caused by fixed sensitivity.

[0073] Secondly, the dynamic graph continuous flow uses known physiological regulatory relationships as graph edges and plant physiological indicators as graph nodes. The adjacency matrix is ​​updated with the low-temperature treatment day counter, which can capture the topological changes of the plant's physiological regulatory network at different developmental stages. Stem flow rate and leaf water potential are used as graph node features to directly participate in graph convolution operations, so that the water stress state is encoded into the physiological state feature vector in a structured way, rather than just as a simple threshold trigger signal.

[0074] Furthermore, the cross-modal attention fusion layer calculates attention weights for the rhythmic pulse feature vector and the physiological state feature vector respectively, and then performs weighted fusion to make the temporal rhythm features and physiological structure features complementary. The differentiation prediction head and the temperature control strategy head simultaneously output the predicted value of flowering probability and the target temperature control parameters based on the fused feature vector. When the jump mechanism detects an abnormal rhythmic pulse pattern, it directly bypasses the temperature control strategy head, activates the emergency protection control sub-network, and outputs the temperature rise rate and water replenishment amount, thereby achieving a protective response under plant stress.

[0075] Finally, the energy-saving flowering game optimization model establishes coupling constraints between flowering rate and cooling energy consumption through a two-layer game structure. It uses a Gaussian process regression surrogate model to replace real simulation evaluation, and a reference-point-guided non-dominated sorting genetic algorithm to quickly solve the Pareto front on the surrogate surface. The optimization results are periodically written back to the differentiated diurnal temperature control scheme, forming a continuous optimization closed loop for temperature control parameters. The coordinated operation of these modules enables temperature control decisions to perceive both the temporal characteristics of the plant's diurnal rhythm and the physiological structural characteristics of the coupled stress of stem flow and water potential. This fundamentally solves the problem of the disconnect between existing temperature control strategies and the actual physiological needs of plants.

[0076] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0077] The specific implementation of step S01 is as follows: Collect genomic single nucleotide polymorphism (SNP) marker data and historical phenotypic data of 30-50 representative Phalaenopsis orchid strains to construct a strain temperature sensitivity threshold prediction model. Experiments are conducted on each strain under orthogonal combinations of 14, 17, and 21 days of low-temperature treatment and night temperature setpoints of 18℃, 20℃, and 22℃. The flowering rate under each combination is recorded, and the temperature sensitivity threshold for each strain is determined through regression analysis, forming a strain temperature sensitivity threshold database, which serves as the model training label. The temperature sensitivity threshold label includes the minimum effective number of low-temperature treatment days required for each strain. (Day) and nighttime lower limit values (°C).

[0078] The specific implementation of step S02 is as follows: Based on the temperature threshold label, set the night temperature value. The temperature (°C) is set within the range of 18–22°C, and the number of days for low-temperature treatment is set within the range of 14–21 days. The optimal cooling curve planning module determines the safe cooling rate range through a 5×6 gradient experiment matrix, and the cooling rate... Take 0.5 to 3 Transcriptome data were collected from plants at each level to analyze the response relationship between heat shock protein gene expression and flowering rate. The response surface was fitted, and the cooling rate range with a flowering rate higher than 85% and heat shock protein gene expression lower than the baseline value by 20% was taken as the safe range. The experiment was repeated 3 times and the average value was taken to confirm the boundary value.

[0079] The specific implementation of step S03 is as follows: Stem flow rate is collected in real time using a stem flow sensor and a leaf microclimate probe. ( ) and leaf water potential ( A coupled dynamic model of temperature and water potential is established, and the coupled stress index is expressed as follows:

[0080] ;

[0081] In the formula, The coupling stress index is dimensionless. The stem flow rate is the baseline value under normal growth conditions. ), The baseline value of leaf water potential under normal growth conditions ( Both are taken as the historical average of the corresponding indicators under normal growth conditions. Coupling stress trigger threshold. The dimensionless value was obtained by recording the primordium necrosis rate under different combinations of temperature jumps of 4–10°C and initial water deficit conditions. The value was defined as the value corresponding to a primordium necrosis rate exceeding 10%. The value was used as a candidate threshold, and after five batches of repeated experiments, the 95th percentile was taken as the final coupled stress triggering threshold. .when Exceed At that time, the temperature slow-rise program and water replenishment control are triggered in conjunction.

[0082] The specific implementation of step S04 is as follows: The temperature time series data and stem flow rate are... Leaf water potential With porosity ( The input is a circadian rhythm sensing temperature control network. Temperature time-series data is converted into discrete pulse sequences based on a rate-of-change threshold. The rate-of-change threshold is determined by the intersection of the temperature change rate distributions of historical normal cooling events and abnormal stress cooling events, and confirmed through three batches of experimental iterations. The discrete pulse sequence is input into a spiking neural network composed of three layers of leaky integral firing neurons. layer( The formula for updating the membrane potential of a leaky integral-fired neuron is expressed as follows:

[0083] ;

[0084] In the formula, For the first Layer At time 1 neuron The membrane potential (dimensionless). For the first The layer leakage coefficient (dimensionless) was initially set in the range of 0.8 to 0.95. For this neuron at time t The pulse firing status (value 0 or 1). For the first Layer The first neuron to the second Layer The synaptic weights of each neuron (dimensionless) are initialized using sparse initialization, with the sparsity set in the range of 80% to 90%. For the first Layer Bias terms of each neuron (dimensionless). For the first Layer neuron index. The spur fire rule is: when... Exceeding the discharge threshold (Dimensionless) And the membrane potential is reset to 0, otherwise Discharge threshold The empirical value is 1.0. The first layer has 128 neurons, the second layer has 64, and the third layer has 32, outputting a rhythmic pulse feature vector with a dimension of 32. , It is a 32-dimensional real vector.

[0085] Stem flow rate Leaf water potential Pore ​​conductance With steady-state temperature value (°C) is used as a feature of a dynamic graph node, and the node feature column vector is expressed as follows:

[0086] ;

[0087] In the formula, These are dimensionless column vectors of node features. The reference value for porosity conductivity ( ), The steady-state temperature reference value (°C) is used. and All values ​​are historical averages of the corresponding indicators under normal growth conditions. (Dynamic adjacency matrix) (Dimensionless) Depending on the plant's growth and development stages (Days, determined by the cryogenic treatment days counter) Dynamically updated, the update formula is expressed as follows:

[0088] ;

[0089] In the formula, The initial known physiological regulatory relationships correspond to the basic adjacency matrix (dimensionless). For the first The number of days corresponding to each growth and development stage transition point. This is a discrete unit impulse function, taking the value 1 when the input is 0, and 0 otherwise. For the first The incremental update matrix (dimensionless) of the adjacency matrix during each stage switch. This represents the total number of stage transitions. For stage switching indexes. The 3rd layer graph convolutional network... The formula for updating the features of layer nodes is expressed as follows:

[0090] ;

[0091] In the formula, To add a self-loop adjacency matrix (dimensionless). It is the identity matrix. for The degree matrix (dimensionless). For the first Layer node feature matrix (dimensionless). For the first Layer-learnable weight matrix (dimensionless). For nonlinear activation functions, empirically, rectified linear units are chosen. For the first The layer outputs a dimensionless feature matrix of nodes. The first hidden layer has a dimension of 64, the second layer has a dimension of 32, the third layer has a dimension of 16, and the output is a physiological state feature vector with a dimension of 16. , It is a 16-dimensional real vector.

[0092] Rhythmic pulse feature vector Physiological state feature vector The input is a cross-modal attention fusion layer with 4 attention heads. One attention point ( The formula for calculating attention weights is as follows:

[0093] ;

[0094] ;

[0095] In the formula, and The first A head of attention and The calculated attention weights (dimensionless) and , For the first The query vector (dimensionless) of each attention head is obtained by linear projection of the fused input. and They are respectively and The corresponding bond vector (dimensionless). The key vector dimension is empirically set to 16. This represents the vector transpose. The fused output is a 64-dimensional fused feature vector. ( (where the vector is a 64-dimensional real number vector), the fusion process formula is expressed as follows:

[0096] ;

[0097] In the formula, and They are respectively and In the The value vector (dimensionless) under each attention head. The fusion layer can learn weight matrix (dimensionless). This represents a vector concatenation operation. It fuses feature vectors. Simultaneously input the differentiation prediction head (outputs the predicted flowering probability value) (dimensionless, value 0-1) and temperature control strategy head (outputs target night temperature setpoint) and target cooling rate When an abnormal rhythmic pulse pattern is detected, the temperature control strategy head is skipped, and the emergency protection control sub-network is activated. Input, output temperature rise rate ( ) and water replenishment ( ).

[0098] Rhythm regulation function coupled with stress index Flowering probability prediction value The rhythm regulation value is calculated by taking the normalized value of the effective cumulative number of low-temperature days as input, and the formula is expressed as follows:

[0099] ;

[0100] In the formula, This is the rhythm regulation value (dimensionless). This represents the current cumulative number of days with effective low temperatures. , , The weighting coefficients and The weight coefficients are obtained by: using the training set... Using the Pearson correlation coefficient with flower bud differentiation results as the criterion, the results were iterated in the range of 0 to 1 with a step size of 0.1. , , The optimal combination was determined by repeating the experiment three times and taking the average of the results. At that time, Lower it to 0.80-0.85; when At that time, maintain Between 0.85 and 0.90; when At that time, Adjusted to 0.90-0.95.

[0101] The specific implementation of step S05 is as follows: An energy-saving flowering game optimization model is constructed with flowering rate and cooling energy consumption as dual objectives. The formula for the upper-level objective function is expressed as follows:

[0102] ;

[0103] The formula for the lower-level objective function is as follows:

[0104] ;

[0105] In the formula, This is the target value for the upper layer (dimensionless). The target value for the lower level is dimensionless. To predict the flowering rate (%) This represents the highest flowering rate (%) in the history of this variety. For cooling energy consumption ( ), Baseline cooling energy consumption ( ), The upper-layer coupling weight coefficient (dimensionless). The lower-level coupling weight coefficients are dimensionless, and the constraints are as follows: ℃ and , The minimum flowering rate (%) required for production is usually taken as 70%. and The method for obtaining the Pareto front is as follows: Iterate through combinations within the range of 0.1 to 0.5 with a step size of 5%, using the maximum hypervolume index of the Pareto front on the validation set as the criterion, repeating the experiment three times and taking the average for confirmation. A Gaussian process regression surrogate model is used to replace the actual simulation evaluation. On the surrogate surface, a non-dominated sorting genetic algorithm guided by reference points is used to quickly approximate the Pareto front, and the Pareto front is then determined. and Write back to the differentiated day and night temperature control scheme.

[0106] The specific implementation of step S06 is as follows: Based on the temperature sensing threshold label and the real-time collected stem flow rate, leaf water potential, stomatal conductance, and temperature time series data, the weight parameters of the diurnal rhythm sensing temperature control network are periodically updated. Network training is divided into three stages: First, the pulse-coded rhythmic flow is pre-trained unsupervised using time-dependent synaptic plasticity rules. The time-dependent synaptic plasticity rules are based on the time difference between pre- and post-synaptic neuron pulse firing. ( The synaptic weights are adjusted based on this, and the weight adjustment formula is expressed as follows:

[0107] ;

[0108] In the formula, For the first Layer synaptic weight The adjustment amount (dimensionless). To enhance the learning rate (dimensionless), an empirical value of 0.01 is recommended. To suppress the learning rate (dimensionless), an empirical value of 0.0105 is used. To enhance the time constant ( ), experience value 20, To suppress the time constant ( ), experience value 20, The timing of postsynaptic neuron pulse firing ( ), The timing of presynaptic neuron pulse firing ( The training iterations consist of 100-200 epochs; then, supervised learning pre-training is applied to the dynamic graph convolutional network, using cross-entropy loss as the loss function, for 50-100 epochs; finally, joint fine-tuning is performed on the entire network, using differentiation prediction loss as the loss function. (Dimensionless) and temperature control strategy loss The (dimensionless) weighted sum is expressed by the following formula:

[0109] ;

[0110] In the formula, Total loss (dimensionless). The weighting coefficients (dimensionless) are determined by searching within the range of 0.3 to 0.7 based on the principle of optimal flower rate in the validation ensemble. The optimizer uses an adaptive moment estimation algorithm, with the initial learning rate set between 0.0005 and 0.002, and 100 to 300 training epochs to complete closed-loop adaptive regulation.

[0111] To better understand and implement this invention, the following is a specific application scenario of the invention, Example 2: Technicians set up a parallel flowering test environment for multiple Phalaenopsis orchid varieties, selected 8 commercially cultivated varieties as test objects, and labeled them as variety A to variety H respectively. The day-night temperature difference induction experiment was carried out simultaneously in the same greenhouse, and the low temperature induction period was set to 21 days. The entire process was controlled by the regulation method of this invention.

[0112] In step S01, technicians extracted genome samples from each of the eight cultivars, obtained single nucleotide polymorphism (SNP) marker data for each cultivar, and constructed a cultivar temperature sensitivity threshold prediction model by combining historical flowering rate records. After the model training was completed, a temperature sensitivity threshold label was generated for each cultivar, specifically including the minimum effective number of days of low-temperature treatment and the lower limit of nighttime temperature, as shown in Table 1.

[0113] Table 1 Temperature Sensing Threshold Labels for Each Variety

[0114]

[0115] Based on the temperature threshold labels shown in Table 1, technicians independently developed differentiated day-night temperature control schemes for each strain in step S02. The target nighttime temperature setting for strains A, D, and G was 18–20°C, and the initial cooling rate was set at 1.5. The target nighttime temperature for strains C and F is set at 18–22°C, and the initial cooling rate is set at 1. The optimal cooling curve planning module generates parameterized cooling paths for each product line based on the temperature threshold labels of each product.

[0116] In step S03, the stem flow sensor and leaf microclimate probe collected stem flow rate, leaf water potential, and stomatal conductance data once per hour. On day 7 of low-temperature induction, the stem flow rate of strain C was... Drop to baseline value 62% of the leaf water potential Drop to baseline value The coupling stress index was calculated to be 71%. The value is 0.44. Based on the coupled stress triggering threshold determined from five batches of experiments (95th percentile, corresponding to a value of 0.40), the system determines that the coupled stress index of strain C exceeds the triggering threshold, and triggers the temperature slow-rise program, setting the nighttime temperature slow-rise rate to 0.5. Simultaneously, water replenishment control was initiated, with 120 mL of water added to strain C plants to protect flower bud primordia from water stress damage. The coupled stress index of the other 7 strains did not exceed the trigger threshold, and the original differentiated temperature control plan continued to be implemented.

[0117] In step S04, the circadian rhythm sensing temperature control network performs parallel inference for the eight varieties. The dynamic graph continuous stream independently maintains the graph topology of each variety using a counter for the number of days of low-temperature treatment as an index. The adjacency matrix is ​​stored in blocks to achieve parallel inference. On the 14th day of low-temperature induction, the circadian rhythm sensing temperature control network outputs the predicted flowering probability values ​​for each variety, as shown in Table 2.

[0118] Table 2. Predicted flowering probability of each variety on day 14 after low-temperature induction.

[0119]

[0120] Based on the predicted flowering probability and target temperature control parameters shown in Table 2, the diurnal rhythm sensing temperature control network writes the target nighttime temperature setpoint and target cooling rate into the corresponding temperature control execution module of each cultivar, realizing differentiated temperature control command output with cultivar-level isolation. The rhythm adjustment function calculates the rhythm adjustment value based on the coupling stress index (0.44), predicted flowering probability (0.72), and normalized value of effective low-temperature cumulative days (14 / 21≈0.67) of cultivar C, and adjusts the leakage coefficient of cultivar C to 0.88, ensuring that the pulse-coded rhythm flow maintains appropriate sensitivity to subsequent temperature change events. Figure 2 As shown, the coordinated change trend of the leakage coefficient adjustment trajectory and the predicted flowering probability value of strain C after the emergency protection is triggered is clearly visible.

[0121] In step S05, the energy-saving flowering game optimization model solves for the Pareto front and upper-level coupling weight coefficients on the surrogate surface for each of the eight varieties. Set to 0.25, lower layer coupling weight coefficient The value is set to 0.30, with the constraint that the flowering rate should not be lower than the historical minimum production requirement for each variety. A Gaussian process regression surrogate model fits the objective function to an approximate surface using simulation samples accumulated during the low-temperature induction period. A reference-point-guided non-dominated sorting genetic algorithm quickly converges to the Pareto front on the surrogate surface. The optimized nighttime temperature setpoint and cooling rate are written back to the differentiated day-night temperature control scheme, such as... Figure 3 As shown, the distribution of each strain on the Pareto front reflects the trade-off between flowering rate and refrigeration energy consumption.

[0122] In step S06, technicians performed a weight update on the diurnal rhythm sensing temperature control network on days 7, 14, and 21 of low-temperature induction. Each update used the currently accumulated stem flow rate, leaf water potential, stomatal conductance, and temperature time-series data as input. An adaptive moment estimation algorithm was used to fine-tune the dynamic graph continuous flow and the entire network. The weighting coefficient of the loss function was set to 0.5, and the initial learning rate was set to 0.001. After three periodic weight updates, the mean square error between the predicted and actual flowering probabilities of each variety continued to decrease, and the accuracy of the diurnal rhythm sensing temperature control network in sensing the physiological state of each variety gradually improved. The closed-loop adaptive regulation effect is shown in Table 3.

[0123] Table 3. Mean Square Error of Flowering Probability Prediction After Weight Update for Each Period

[0124]

[0125] like Figure 3 As shown, the hypervolume index of the Pareto front improved after each weight update, indicating that the energy-saving flowering game optimization model continuously improves its approximation accuracy of the real target surface as the surrogate model samples accumulate. The actual flowering situation of the eight varieties after the low-temperature induction period is shown in Table 4.

[0126] Table 4. Statistics on actual flowering conditions of each variety

[0127]

[0128] As shown in Table 4, strain C completed the entire 21-day effective low-temperature accumulation even after the emergency protection was triggered, with an actual flowering rate of 78%. This indicates that the intervention of the emergency protection control subnetwork effectively prevented flower bud primordia necrosis, and the low-temperature induction process was not interrupted. The other strains all completed the low-temperature induction cycle normally according to the differentiated diurnal temperature control scheme, with flowering rates ranging from 71% to 94%, which is highly consistent with the flowering probability prediction value output by the diurnal rhythm sensing temperature control network on day 14.

[0129] Compared to traditional fixed-parameter temperature control methods, this invention represents a fundamental advancement in technical principles. Traditional methods rely on static rule bases or lookup table systems, where the generation logic of temperature control commands is completely decoupled from the real-time physiological state of the plant. Regardless of the plant's current level of water stress, the system executes cooling operations according to preset parameters, and the risk of flower bud primordia necrosis relies entirely on post-event manual intervention. This invention, through heterogeneous dual-flow modeling of pulse-coded rhythmic flow and dynamic graph continuous flow, jointly encodes the plant's rhythmic temporal characteristics and physiological structural characteristics into the same decision framework. This establishes a causal relationship between the generation process of temperature control commands and the plant's physiological state, rather than a simple threshold triggering relationship. The rhythm regulation function further feeds back the coupled stress state to the dynamic sensitivity adjustment of the pulse-coded rhythmic flow, forming an internal closed loop from physiological perception to decision sensitivity. This allows the system to automatically improve the response accuracy to temperature pulse events when the plant's stress state intensifies, and automatically reduce the probability of false triggering when the plant's state is stable. The energy-saving flowering game optimization model combines surrogate surfaces and multi-objective evolutionary algorithms to transform temperature control parameters, which rely on manual experience in traditional methods, into a Pareto optimal solution set that can be solved in real time. This allows the temperature control strategy to achieve a quantitative trade-off between ensuring flowering rate and energy consumption constraints, rather than relying on the subjective judgment of operators. The synergistic effect of these mechanisms enables this invention to possess, in multi-variety parallel flowering scenarios, the variety-level isolation decision-making capability and real-time physiological response capability that are impossible to achieve at the principle level by traditional methods.

[0130] It should be noted that the variables involved in this invention are explained in detail in Tables 5 and 6.

[0131] Table 5. Variable Explanation Table (Part 1)

[0132]

[0133] Table 6. Variable Explanation Table (Part Two)

[0134]

[0135] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for regulating Phalaenopsis flower bud differentiation by utilizing diurnal temperature variation, characterized in that, Includes the following steps: We collected genomic single nucleotide polymorphism markers and historical phenotypic data of Phalaenopsis orchids from various strains, constructed a strain temperature sensitivity threshold prediction model, and generated a temperature sensitivity threshold label for each strain. Based on the temperature threshold label, a differentiated day and night temperature control scheme is formulated, and an optimal cooling curve planning module is embedded to control the cooling rate within the cooling rate range. During the diurnal temperature range induction period, stem flow sensor and leaf microclimate probe are used to collect stem flow rate and leaf water potential in real time, and a temperature-water potential coupled dynamic model is established. When the coupled stress index exceeds the coupled stress triggering threshold, the temperature slow rise program and water replenishment control are triggered in conjunction. Temperature time series data, stem flow rate, leaf water potential and stomatal conductance are input into the diurnal rhythm sensing temperature control network. The diurnal rhythm sensing temperature control network outputs the predicted value of flowering probability, the target night temperature setpoint and the target cooling rate in real time. The target night temperature setpoint and the target cooling rate are written into the temperature control execution module. With flowering rate and cooling energy consumption as dual objectives, an energy-saving flowering game optimization model is constructed. The Pareto front is solved on the surrogate surface, and the night temperature setpoint and cooling rate corresponding to the Pareto front are written back to the differentiated day and night temperature control scheme. Based on the temperature threshold label and real-time collected stem flow rate, leaf water potential, stomatal conductance and temperature time series data, the weight parameters of the diurnal rhythm sensing temperature control network are periodically updated to complete closed-loop adaptive regulation.

2. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 1, characterized in that, The acquisition of the temperature sensitivity threshold labels involves conducting orthogonal experiments on representative varieties under multiple low-temperature treatment days and multiple night temperature setpoints, recording the flowering rate, determining the temperature sensitivity threshold of each variety through regression analysis, forming a variety temperature sensitivity threshold database, and using the variety temperature sensitivity threshold database as the training label for the variety temperature sensitivity threshold prediction model.

3. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 2, characterized in that, The cooling rate range of the optimal cooling curve planning module is specifically determined by collecting transcriptome data from plants at each cooling rate level through gradient cooling experiments, analyzing the response relationship between heat shock protein gene expression and flowering rate, fitting the response surface between cooling rate and flowering rate, and taking the cooling rate range where the flowering rate is higher than the flowering rate threshold and the heat shock protein gene expression is lower than the baseline value by a certain proportion as the safe cooling rate range.

4. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 3, characterized in that, The coupled stress index is calculated from stem flow rate and leaf water potential. Specifically, it is obtained by multiplying the ratio of real-time stem flow rate to stem flow rate baseline by the ratio of real-time leaf water potential to leaf water potential baseline.

5. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 4, characterized in that, The coupling stress triggering threshold is obtained by recording the necrosis rate of flower bud primordia under different combinations of temperature jump and initial water deficit states. The coupling stress index value corresponding to the flower bud primordia necrosis rate exceeding the necrosis rate threshold is used as the quantile of the coupling stress triggering threshold after multiple batches of repeated experiments.

6. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 5, characterized in that, The diurnal rhythm sensing temperature control network includes a pulse-coded rhythmic stream, a dynamic graph continuous stream, and a cross-modal attention fusion layer. Temperature time-series data is converted into discrete pulse sequences according to the rate of change threshold and then input into the pulse-coded rhythmic stream. Stem flow rate, leaf water potential, stomatal conductance, and steady-state temperature value are used as dynamic graph node features and input into the dynamic graph continuous stream. The two outputs are fused by the cross-modal attention fusion layer and then input into the differentiation prediction head and the temperature control strategy head, respectively.

7. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 6, characterized in that, The pulse-coded rhythm stream is composed of a stack of multiple leaky integral firing neurons. The interlayers are pre-trained using time-dependent synaptic plasticity rules, and the interlayer synaptic weight matrix is ​​sparsely initialized.

8. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 7, characterized in that, The circadian rhythm sensing temperature control network introduces a rhythm adjustment function, which takes the coupling stress index, the predicted flowering probability, and the normalized value of the effective cumulative number of low temperature days as inputs to calculate the rhythm adjustment value. Based on the rhythm adjustment value, the leakage coefficient of the leaking integral firing neurons in the pulse-coded rhythm stream is dynamically adjusted.

9. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 8, characterized in that, The circadian rhythm sensing temperature control network is equipped with a jump mechanism. When an abnormal circadian rhythm pulse pattern is detected, the normal output of the temperature control strategy head is skipped, and the emergency protection control sub-network is directly activated to output the temperature rise rate and water replenishment amount.

10. The method for regulating Phalaenopsis orchid bud differentiation by utilizing diurnal temperature variation according to claim 9, characterized in that, The dynamic graph continuous flow is composed of a multi-layer graph convolutional network. The adjacency matrix of the dynamic graph is dynamically updated according to the plant growth and development stage, which is determined by a counter for the number of days of low-temperature treatment.

Images