Dynamic adjusting method for concentration of hydrogen-rich water

Abstract

The invention provides a hydrogen-rich water concentration dynamic adjustment method, and relates to the field of concentration adaptive adjustment, and the method comprises the following steps: S1, data acquisition; step S2, data preprocessing; s3, key influence factors are extracted; s4, acquiring the initial concentration of the hydrogen-rich water; and S5, obtaining the optimized concentration of the hydrogen-rich water: obtaining a dynamic concentration adjustment amount, an extreme environment adjustment amount and an implicit stress adjustment amount through reinforcement learning dynamic optimization, extreme environment adaptation and implicit stress early warning, and obtaining the final concentration of the hydrogen-rich water. According to the method, self-adaptive adjustment of the concentration of the hydrogen-rich water can be performed according to the growth state of the plant and environmental factors, the accuracy of the concentration of the applied hydrogen-rich water is ensured, and the problem that normal growth of the plant is affected due to too high or too low concentration of the hydrogen-rich water is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of concentration adaptive regulation, and in particular to a method for dynamically regulating the concentration of hydrogen-rich water. Background Art

[0002] Hydrogen-rich water, or water containing high concentrations of dissolved hydrogen, is increasingly being used in agricultural plantings with the advancement of science and technology and the development of modern agriculture. Adding hydrogen-rich water to irrigation water utilizes the antioxidant properties of hydrogen in the water to regulate plant physiological metabolism. This selectively neutralizes excess reactive oxygen species (ROS, such as hydroxyl radicals and superoxide anions) produced during plant growth, reducing oxidative damage and protecting the functions of organelles like chloroplasts and mitochondria. Hydrogen also promotes the synthesis of growth-promoting hormones such as auxin (IAA) and cytokinin (CTK) in plants, while inhibiting the accumulation of stress hormones like abscisic acid (ABA), effectively regulating plant cell division, growth, and biosynthesis. Furthermore, hydrogen-rich water can regulate soil microbial communities, improve soil aeration, and promote root development, thereby enhancing plants' ability to absorb water and nutrients, and improving their survival and growth. Although hydrogen-rich water can effectively promote plant growth and increase agricultural production and efficiency, if the concentration of hydrogen-rich water used for irrigation is too high, that is, the concentration of hydrogen-rich water does not match the growth status of the plant, it will not only disrupt the redox balance in the plant (that is, excessively high concentrations of hydrogen-rich water will excessively remove ROS in the plant, affecting cell division and gene expression, leading to stagnation of root development and hindering pollen germination), but also inhibit plant growth and development (for example: decreased activity of the root apical meristem, shortened root length, and reduced number of root hairs; inhibited stem elongation, shortened internodes, smaller leaves and blocked chlorophyll synthesis; early or delayed flowering, accelerated petal shedding, and deformed fruit development, etc.), reduce photosynthesis efficiency (photosynthesis depends on the stability of chloroplast structure and the activity of photosynthetic enzymes, and high concentrations of hydrogen will destroy the thylakoid membrane structure, inhibit the activity of key enzymes for photosynthetic carbon assimilation, and cause abnormal closure of leaf stomata), and lead to an imbalance of plant nutrients (high concentrations of hydrogen will cause soluble sugars and vitamin C to deplete The content of beneficial ingredients such as oxalic acid and nitrite will decrease, while harmful substances such as oxalic acid and nitrite will accumulate).

[0003] In the existing agricultural planting process, the concentration of hydrogen-rich water is usually adjusted based on the operator's past experience and a large number of experiments. The human influence factor is high, and it is very easy to have problems such as the hydrogen-rich water concentration being too high or too low. If the concentration is too high, it will cause large-scale plant withering or yield reduction. If the concentration is too low, the purpose of applying hydrogen-rich water cannot be achieved, and repeated and multiple applications are required, which is time-consuming and labor-intensive. It can be seen that the existing technology lacks a method for adaptively adjusting the concentration of hydrogen-rich water according to the growth trend of the plant, which seriously restricts the application of hydrogen-rich water in agricultural planting. Summary of the Invention

[0004] In response to the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method for dynamically adjusting the concentration of hydrogen-rich water. This method can adaptively adjust the concentration of hydrogen-rich water according to the growth status of the plant and environmental factors, ensure the accuracy of the applied hydrogen-rich water concentration, and avoid the problem of excessively high or low hydrogen-rich water concentration.

[0005] The purpose of the present invention is achieved through the following technical solutions: A method for dynamically adjusting the concentration of hydrogen-rich water, comprising: Step S1, data collection: deploying data collection sensors based on crop types and field conditions to collect plant growth trend data and field environmental factor data respectively; Step S2, data preprocessing: performing noise removal, abnormal data detection and correction, and data normalization preprocessing on the collected data; Step S3, extraction of key influencing factors: exploring plant growth trend data and field environmental factor data to determine the key factors affecting crop growth; Step S4, obtaining the initial concentration of hydrogen-rich water: inputting the key influencing factor data into the fuzzy logic system, defining the fuzzy subsets of each key influencing factor, and obtaining the initial concentration of hydrogen-rich water; Step S5, obtaining the optimized concentration of hydrogen-rich water: obtaining the dynamic concentration adjustment amount, extreme environment adjustment amount, and implicit stress adjustment amount through reinforcement learning dynamic optimization, extreme environment adaptation, and implicit stress early warning, respectively, to obtain the final concentration of hydrogen-rich water.

[0006] Based on further optimization of the above scheme, the growth trend data include chlorophyll fluorescence, canopy multispectral, stem micro-deformation degree, leaf infrared temperature, root growth, etc.; field environmental factor data include light intensity, air temperature and humidity, soil temperature and humidity, CO2 concentration, soil pH value, etc.

[0007] Based on further optimization of the above solution, the noise removal in step S2 is to use a Kalman filter algorithm to remove random noise.

[0008] Based on further optimization of the above solution, the abnormal data detection and correction in step S2 is specifically as follows: For input feature data X =[ x 1, x 2,…, x n ], the isolation forest algorithm is used to detect anomalies in the data after random noise is removed, and the data anomaly score is obtained:

[0009] Where: h(X) Represents a sample X path length in the isolation tree;E[h(X)] represents the expected path length; represents Euler's constant; Preset anomaly score threshold SX d ,like , then the corresponding sample is determined to be an outlier; For outliers, Huber robust regression is used for correction, and the Huber loss function is:

[0010]

[0011] Where: is a constant, generally 1.35; by optimizing the model parameters Improve fault tolerance to extreme environments.

[0012] Based on further optimization of the above scheme, the key influencing factor extraction in step S3 is specifically as follows: based on the key point identification of the "stress-metabolism" coupling network, the correlation analysis of dynamic time lag, the key variable allocation of multi-scale feature fusion, and the stress state adaptive dynamic key factors, the key factors are extracted: that is, if a variable is identified as a key influencing factor by at least three of the four methods of key point identification based on the "stress-metabolism" coupling network, the correlation analysis of dynamic time lag, the key variable allocation of multi-scale feature fusion, and the stress state adaptive dynamic key factors, then the variable is a key influencing factor; if a variable is identified as a key influencing factor by only one of the four methods of key point identification based on the "stress-metabolism" coupling network, the correlation analysis of dynamic time lag, the key variable allocation of multi-scale feature fusion, and the stress state adaptive dynamic key factors, then it is directly excluded; if a variable is identified as a key influencing factor by any two of the four methods of key point identification based on the "stress-metabolism" coupling network, the correlation analysis of dynamic time lag, the key variable allocation of multi-scale feature fusion, and the stress state adaptive dynamic key factors, then it is used as a secondary influencing factor for concentration correction.

[0013] Based on the further optimization of the above scheme, the key point identification based on the "stress-metabolism" coupling network is as follows: a model for estimating the level of reactive oxygen species (ROS) is established:

[0014] Where: T leaf Indicates the infrared temperature of the leaves, T air Indicates air temperature; represents the volumetric moisture content of the soil, represents field capacity; F v / F m represents the chlorophyll fluorescence parameter; K 1. K 2. K 3 represents the corresponding weight coefficient; represents the error term; Based on the reactive oxygen species level estimation model and the partial least squares path model (PLS-PM), the preprocessed data variables were divided into three layers: the environmental layer (CeX) of the ROS generation driver, the physiological layer (CeY) of the ROS damage target, and the effect layer (CeZ) of the ROS regulation result. Based on the mediating role of ROS, the paths CeX→CeY and CeY→CeZ were established, and the path coefficients of the two paths were obtained: 、 ; Preset path coefficient threshold ,like , then the corresponding variables are taken as key influencing factors; Dynamic time-lag correlation analysis: Quantifying the dynamic correlation between variables using time-lag mutual information: Growth trends or field environmental factor variables X(t) and concentration requirements , in the time lag Mutual information under TDMI for:

[0015] Where: P represents the probability distribution, This means that the variable X is at time t Pick a Value and concentration demand B at time Pick b The probability of the value; Calculate the TDMI hysteresis of each growth trend or field environmental factor variable separately. The variable corresponding to the peak value of TDMI is taken as the key influencing factor; Key variable allocation for multi-scale feature fusion: For time series variables X(t) Perform three-layer wavelet packet decomposition (i.e., instantaneous (hourly) - short-term (dayly) - long-term (weekly); instantaneous response: sudden rise in leaf temperature (direct sunlight) → hydrogen-rich water needs to quickly alleviate transpiration stress; short-term response: slight deformation of stems (water deficit) → hydrogen-rich water needs to continuously adjust turgor pressure; long-term response: root growth (nutrient absorption) → hydrogen-rich water needs to promote metabolic repair), and obtain the energy values ​​of eight frequency bands. E i,j :

[0016] Where: i Indicates the variable number; j Indicates the frequency band number; represents the wavelet basis function; The sub-signal of each frequency band is input into the random forest as an independent feature to obtain the Gini importance of each sub-signal; based on the metabolic cycle of hydrogen-rich water, different frequency bands are given weights. , and obtain the comprehensive importance of the corresponding variables:

[0017] Where: Gini i,j Indicates the i The variable j Gini importance of frequency band features; Preset comprehensive importance threshold I d ,If the comprehensive importance of a variable is greater than the comprehensive importance threshold, the corresponding variable is regarded as a key influencing factor; Key factors for adaptive dynamics of stress states: Introduction of stress index SI:

[0018] Where: T opt Indicates the most suitable leaf temperature; T max Indicates critical high temperature; K 4. K 5 respectively represent the corresponding weight coefficients; Preset stress index threshold SI d : when When , it means that the plant metabolism is basically normal. The posterior probability of the variables and the concentration of hydrogen-rich water is calculated through the Bayesian network. The variables are arranged from large to small and the top three variables are taken as the key influencing factors (for example: canopy multispectral, light intensity, CO2 concentration); when When , it means that the photosynthesis / respiration of the plant is inhibited. The posterior probability of the variables and the concentration of hydrogen-rich water is recalculated through the Bayesian network, and the variables are arranged from large to small and the top three variables are taken as key influencing factors (for example: the degree of micro-deformation of the stem, infrared temperature of the leaves, and soil temperature and humidity).

[0019] Based on further optimization of the above solution, step S4 is specifically as follows: Define the membership function of the fuzzy subset of key influencing factors, including a small number of membership U dr (S), appropriate membership U su (S) Excess membership U we (S) :

[0020] Where: S represents the current measured value of the key influencing factor; S min Indicates the critical value when the key influencing factor is too low; S a Indicates the appropriate central value of the key influencing factors; S max Indicates the critical value when the key influencing factor is too high; The fuzzy rule base is constructed by integrating multi-factor association and expert practical experience, and the center of gravity method is used to defuzzify and obtain the preliminary concentration:

[0021] Where: U i Indicates the i The membership degree of the rule; C i Indicates the i Reference concentration for the rule; m Indicates the number of activated rules.

[0022] Based on the further optimization of the above solution, the reinforcement learning dynamic optimization is specifically as follows: The initial concentration and the key influencing factor vector of real-time monitoring are taken as the state space Y, and the concentration adjustment of hydrogen-rich water is taken as the action space A to construct the reward function:

[0023] Where: Indicates the change in photosynthetic efficiency; Indicates the change in root growth; Indicates the adjustment range of hydrogen-rich water concentration; Respectively represent the corresponding weight coefficients; Update the Q value through the Q-Learning algorithm:

[0024] Where: represents the learning rate; represents the discount factor; Indicates a new state The maximum Q value of all possible actions under When the Q value change is less than the preset threshold, the Q value table converges; after the Q value table converges, for any current state Y , select the action that maximizes the Q value A * :

[0025] action A * That is, the adjustment amount of dynamic optimization of hydrogen-rich water concentration reinforcement learning under the current state .

[0026] Based on the further optimization of the above solution, the extreme environment adaptation is specifically as follows: Collect historical extreme environment data and build a meta-task set:

[0027] Where: A small amount of training data representing the kth extreme scenario; Represents the test data of the kth extreme scenario; Use MAML (Model-Agnostic Meta-Learning) to pre-train common initial parameters : For each meta-task k ,use Data is updated within the task:

[0028] Where: represents the inner loop learning rate; represents the training loss of task k; Represents the gradient operator, that is, the parameter Find partial derivatives; use Data is meta-updated and initial parameters are optimized:

[0029] Where: represents the outer loop learning rate; Isolation forest is used to detect the novelty of the environment state, output a novelty score, and preset a novelty score threshold. If the novelty score is greater than the novelty score threshold, it is determined to be a new extreme environment. In the new extreme environment, real-time data is collected and meta-learning updates are performed:

[0030] Where: Represents the loss function of new environment data; Use the adapted parameters Combined with reinforcement learning dynamic optimization to obtain Q(Y,A; )The biggest move A * :

[0031] That is, to obtain the adjustment amount of hydrogen-rich water concentration to adapt to extreme environments .

[0032] Based on the further optimization of the above solution, the implicit coercion warning is specifically as follows: Extracting microscopic features of early latent stress from chlorophyll fluorescence and Raman spectroscopy data and constructing stress feature vectors Z x ; Variational Autoencoder (VAE) is used to learn the feature distribution under normal conditions to obtain early implicit coercion: the VAE encoder converts the feature Z x Mapping to latent space , output mean and variance; the VAE decoder converts the latent space variables Refactoring into features (in, and Z x The dimensions are the same), the loss function of VAE is:

[0033] Where: Represents the "conditional probability distribution" defined by the decoder, that is, the description of the reconstructed features is When , the probability distribution of the original input feature Z; represents the expectation of reconstruction loss; represents the approximate posterior distribution of the encoder output; represents the prior distribution of the latent variable; represents KL divergence; represents the balance coefficient; Calculate the reconstruction error REW:

[0034] Where: Represents the feature vector Z x Dimensions; Preset error threshold ,If the reconstruction error REW is greater than the error threshold, it is determined to be early implicit coercion; For the current feature Z now , calculate its difference with the historical coercion library { Z 1,Z 2,…, Z n}'s Euclidean distance:

[0035] Select the one with the smallest distance h samples to obtain the initial adjustment amount:

[0036] Where: C h,i Indicates the historical adjustment amount corresponding to the minimum distance sample; Combined with reinforcement learning dynamic optimization, the action space A in reinforcement learning dynamic optimization is For the fine-tuning interval centered on , the reward function is updated as:

[0037] Implicit coercion adjustment for:

[0038] Where: Represents the gradient of the Q value to the reward function, reflecting the impact of the change in the adjustment amount on the reward; Represents the fine-tuning coefficient.

[0039] The following are the technical effects of the technical solution of the present invention: The present invention extracts key influencing factors from the data of plant growth trends and field environmental factors that affect hydrogen-rich water concentration through four methods: key point identification based on the "stress-metabolism" coupling network, dynamic time-delayed correlation analysis, key variable allocation of multi-scale feature fusion, and adaptive dynamic key factors of stress state. Through the combination of ROS coupling model, time-delayed mutual information, wavelet fusion, and adaptive stress switching, the limitations of traditional static analysis are broken through, and the key influencing factors that are strongly associated with the hydrogen-rich water concentration demand are finally determined, thereby achieving accurate monitoring and regulation, abandoning redundant factor detection, and improving monitoring and control efficiency. At the same time, the present invention uses fuzzy logic to process the nonlinear fuzzy relationship between factors, converts the precise input into fuzzy variables through the membership function, and then realizes multi-factor coupling decision-making through the fuzzy rule base, thereby accurately capturing the multi-factor nonlinear coupling of the actual farmland crop growth process; using reinforcement learning dynamic weight adaptation, the hydrogen-rich water concentration adjustment is regarded as a "real-time interaction process between the intelligent agent and the farmland environment", so that the model can autonomously learn the optimal decision in the environmental changes, effectively solving the problem that the weight coefficient statically set based on historical data cannot respond to sudden changes in the field in real time.

[0040] The present invention can adjust the concentration of hydrogen-rich water in real time, continuously and accurately according to the growth status of the plants and environmental factors, thereby improving the accuracy, adaptability and stability of hydrogen-rich water irrigation, thereby meeting the needs of complex and changeable actual field scenarios, and avoiding problems such as plant yield reduction and death due to excessively high hydrogen-rich water concentration or poor irrigation effect and cost increase due to too low hydrogen-rich water concentration. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a structural block diagram of the dynamic regulation of hydrogen-rich water concentration in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present invention will be described clearly and completely below. In the following description, for the purpose of illustration rather than limitation, specific details such as specific system structures and technologies are proposed to facilitate a thorough understanding of the embodiments of the present invention.

[0043] Example 1: A method for dynamically adjusting the concentration of hydrogen-rich water, comprising: Step S1, data collection: Deploy collection sensors based on crop types and field conditions to collect plant growth trend data and field environmental factor data respectively; growth trend data includes chlorophyll fluorescence, canopy multispectral, stem micro-deformation degree, leaf infrared temperature, root growth, etc.; field environmental factor data includes light intensity, air temperature and humidity, soil temperature and humidity, CO2 concentration, soil pH value, etc.

[0044] Step S2, data preprocessing: The collected data is preprocessed by using a Kalman filter algorithm to remove random noise (the existing conventional Kalman filter can be used), abnormal data detection and correction, and data normalization (the existing conventional normalization method can be used); the abnormal data detection and correction are specifically as follows: For input feature data X =[ x 1, x 2,…, x n ], the isolation forest algorithm is used to detect anomalies in the data after random noise is removed, and the data anomaly score is obtained:

[0045] Where: h(X) Represents a sample X path length in the isolation tree; E[h(X)] represents the expected path length; represents Euler's constant; Preset anomaly score threshold SX d (usually 0.7), if , then the corresponding sample is determined to be an outlier; For outliers, Huber robust regression is used for correction, and the Huber loss function is:

[0046]

[0047] Where: is a constant, generally 1.35; by optimizing the model parameters Improve fault tolerance to extreme environments.

[0048] Step S3, extraction of key influencing factors: Exploring plant growth trend data and field environmental factor data to determine the key factors affecting crop growth; specifically, extraction of key factors is achieved through key point identification based on the "stress-metabolism" coupling network, dynamic time-lag correlation analysis, key variable allocation based on multi-scale feature fusion, and adaptive dynamic key factors based on stress state. If a variable is identified as a key influencing factor by at least three of the four methods, namely, key point identification based on the "stress-metabolism" coupling network, correlation analysis of dynamic time lag, key variable allocation of multi-scale feature fusion, and stress state adaptive dynamic key factors, then the variable is a key influencing factor; if a variable is identified as a key influencing factor by only one of the four methods, namely, key point identification based on the "stress-metabolism" coupling network, correlation analysis of dynamic time lag, key variable allocation of multi-scale feature fusion, and stress state adaptive dynamic key factors, then it is directly excluded; if a variable is identified as a key influencing factor by any two of the four methods, namely, key point identification based on the "stress-metabolism" coupling network, correlation analysis of dynamic time lag, key variable allocation of multi-scale feature fusion, and stress state adaptive dynamic key factors, then it is used as a secondary influencing factor for concentration correction.

[0049] Identification of key points based on the "stress-metabolism" coupling network: Establishment of a reactive oxygen species (ROS) level estimation model:

[0050] Where: T leaf Indicates the infrared temperature of the leaves, T air Indicates air temperature; represents the volumetric moisture content of the soil, represents field capacity; F v / F m represents the chlorophyll fluorescence parameter; K 1. K 2. K3 represents the corresponding weight coefficient; represents the error term; Based on the reactive oxygen species level estimation model and the partial least squares path model (PLS-PM), the preprocessed data variables were divided into three layers: the environmental layer (CeX) of the ROS generation driver, the physiological layer (CeY) of the ROS damage target, and the effect layer (CeZ) of the ROS regulation result. Based on the mediating role of ROS, the paths CeX→CeY and CeY→CeZ were established, and the path coefficients of the two paths were obtained: 、 ; For example, the latent variables of the environmental layer are:

[0051] Where: Indicates the j The weight coefficient of each environmental explicit variable; Indicates that all environmental variables are normalized; Synchronously generate physiological layer latent variables:

[0052] PLS-PM iterates through the cycle of "external weight → internal weight → path coefficient" until the weight change is less than a threshold (e.g., 10 -4 ), stop the iteration and output the final path coefficient; in: External weights:

[0053] Internal weights:

[0054] Where: Cov() represents covariance; Var() represents variance; Similarly, obtain ; Preset path coefficient threshold (usually 0.5), if , then the corresponding variables are taken as key influencing factors; Dynamic time-lag correlation analysis: Quantifying the dynamic correlation between variables using time-lag mutual information: Growth trends or field environmental factor variables X(t) and concentration requirements , in the time lag Mutual information under TDMI for:

[0055] Where: P represents the probability distribution, This means that the variable X is at time t Pick a Value and concentration demand B at time Pick b The probability of the value; Calculate the TDMI hysteresis of each growth trend or field environmental factor variable separately. The change of TDMI (usually 0 to 48 hours) is taken as the key influencing factor. Key variable allocation for multi-scale feature fusion: For time series variables X(t) Perform three-layer wavelet packet decomposition (i.e., instantaneous (hourly) - short-term (dayly) - long-term (weekly); instantaneous response: sudden rise in leaf temperature (direct sunlight) → hydrogen-rich water needs to quickly alleviate transpiration stress; short-term response: slight deformation of stems (water deficit) → hydrogen-rich water needs to continuously adjust turgor pressure; long-term response: root growth (nutrient absorption) → hydrogen-rich water needs to promote metabolic repair), and obtain the energy values ​​of eight frequency bands. E i,j :

[0056] Where: i Indicates the variable number; j Indicates the frequency band number (e.g. j =1, indicating high frequency → hourly level; j =5, indicating medium frequency → day level; j =8, indicating low frequency → cycle level); represents the wavelet basis function; The sub-signal of each frequency band is input into the random forest as an independent feature to obtain the Gini importance of each sub-signal; based on the metabolic cycle of hydrogen-rich water, different frequency bands are given weights. (Generally high frequency , IF , low frequency ), and obtain the comprehensive importance of the corresponding variables:

[0057] Where: Gini i,j Indicates the i The variable j Gini importance of frequency band features; Preset comprehensive importance threshold I d ,If the comprehensive importance of a variable is greater than the comprehensive importance threshold, the corresponding variable is regarded as a key influencing factor; Key factors for adaptive dynamics of stress states: Introduction of stress index SI:

[0058] Where: T opt Indicates the most suitable leaf temperature; T max Indicates critical high temperature; K 4. K 5 respectively represent the corresponding weight coefficients; Preset stress index threshold SI d (usually 0.3): when When , it means that the plant metabolism is basically normal. The posterior probability of the variables and the concentration of hydrogen-rich water is calculated through the Bayesian network. The variables are arranged from large to small and the top three variables are taken as the key influencing factors (for example: canopy multispectral, light intensity, CO2 concentration); when When , it means that the photosynthesis / respiration of the plant is inhibited. The posterior probability of the variables and the concentration of hydrogen-rich water is recalculated through the Bayesian network. The variables are arranged from large to small and the top three variables are taken as the key influencing factors (for example: the degree of stem micro-deformation, leaf infrared temperature, soil temperature and humidity); Among them, the Bayesian network is a probabilistic graphical model that outputs the posterior probability through the causal relationship between "field environmental factors / plant growth trends" and "hydrogen-rich water concentration requirements".

[0059] Step S4, obtaining the initial concentration of hydrogen-rich water: inputting the key influencing factor data into the fuzzy logic system, defining the fuzzy subsets of each key influencing factor, and obtaining the initial concentration of hydrogen-rich water; specifically: Define the membership function of the fuzzy subset of key influencing factors (range [0,1]), including a small number of membership U dr (S) , appropriate membership U su (S) Excess membership U we (S) (Take soil moisture as an example, a small amount of membership is drought membership, and an excessive amount of membership is over-wet membership):

[0060] Where: S represents the current measured value of the key influencing factor; S min Indicates the critical value when the key influencing factor is too low; S aIndicates the appropriate central value of the key influencing factors; S max Indicates the critical value when the key influencing factor is too high; The fuzzy rule base is constructed by integrating multi-factor association and expert practical experience, and the center of gravity method is used to defuzzify and obtain the preliminary concentration:

[0061] Where: U i Indicates the i The membership degree of the rule; C i Indicates the i Reference concentration for the rule; m Indicates the number of activated rules.

[0062] Step S5, obtaining the optimized concentration of hydrogen-rich water: obtaining the dynamic concentration adjustment amount, extreme environment adjustment amount, and implicit stress adjustment amount through reinforcement learning dynamic optimization, extreme environment adaptation, and implicit stress warning, respectively. Reinforcement learning dynamic optimization is specifically as follows: The initial concentration and the key influencing factor vector of real-time monitoring are taken as the state space Y, and the concentration adjustment of hydrogen-rich water is taken as the action space A to construct the reward function:

[0063] Where: Indicates the change in photosynthetic efficiency; Indicates the change in root growth; Indicates the adjustment range of hydrogen-rich water concentration; Respectively represent the corresponding weight coefficients; Update the Q value through the Q-Learning algorithm:

[0064] Where: Indicates the learning rate, generally 0.1 to 0.3; Represents the discount factor, generally 0.7 to 0.9; Indicates a new state The maximum Q value of all possible actions under When the Q value change is less than the preset threshold (usually 0.01), the Q value table converges; after the Q value table converges, for any current state Y , select the action that maximizes the Q value A * :

[0065] action A *That is, the adjustment amount of dynamic optimization of hydrogen-rich water concentration reinforcement learning under the current state .

[0066] Extreme environment adaptation is specifically as follows: Collect historical extreme environment data and build a meta-task set:

[0067] Where: A small amount of training data representing the kth extreme scenario; Represents the test data of the kth extreme scenario; Use MAML (Model-Agnostic Meta-Learning) to pre-train common initial parameters : For each meta-task k ,use Data is updated within the task:

[0068] Where: represents the inner loop learning rate; represents the training loss of task k; Represents the gradient operator, that is, the parameter Find partial derivatives; use Data is meta-updated and initial parameters are optimized:

[0069] Where: represents the outer loop learning rate; Isolation forest is used to detect the novelty of the environment state and output a novelty score. A novelty score threshold is preset (usually 0.9). If the novelty score is greater than the novelty score threshold, it is determined to be a new extreme environment. In the new extreme environment, real-time data is collected and meta-learning updates are performed:

[0070] Where: Represents the loss function of new environment data; Use the adapted parameters Combined with reinforcement learning dynamic optimization to obtain Q(Y,A; )The biggest move A * :

[0071] That is, to obtain the adjustment amount of hydrogen-rich water concentration to adapt to extreme environments .

[0072] The specific warnings for implicit coercion are: Extracting microscopic features of early latent stress from chlorophyll fluorescence and Raman spectroscopy data and constructing stress feature vectors Z x (usually includes 12 chlorophyll fluorescence parameters and three Raman spectral indices, forming a 15-dimensional feature vector); Variational Autoencoder (VAE) is used to learn the feature distribution under normal conditions to obtain early implicit coercion: the VAE encoder converts the feature Z x Mapping to latent space , output mean and variance; the VAE decoder converts the latent space variables Refactoring into features (in, and Z x The dimensions are the same), the loss function of VAE is:

[0073] Where: Represents the "conditional probability distribution" defined by the decoder, that is, the description of the reconstructed features is When , the probability distribution of the original input feature Z; represents the expectation of reconstruction loss; represents the approximate posterior distribution of the encoder output; represents the prior distribution of the latent variable; represents KL divergence; Indicates the balance coefficient (usually 1); Calculate the reconstruction error REW:

[0074] Where: Represents the feature vector Z x Dimensions; Preset error threshold (usually 0.8), if the reconstruction error REW is greater than the error threshold, it is determined to be early implicit coercion; For the current feature Z now , calculate its difference with the historical coercion library { Z 1, Z 2,…, Z n}'s Euclidean distance:

[0075] Select the one with the smallest distance h samples to obtain the initial adjustment amount:

[0076] Where: C h,i Indicates the historical adjustment amount corresponding to the minimum distance sample; Combined with reinforcement learning dynamic optimization, the action space A in reinforcement learning dynamic optimization is For the fine-tuning interval centered on , the reward function is updated as:

[0077] Implicit coercion adjustment for:

[0078] Where: Represents the gradient of the Q value to the reward function, reflecting the impact of the change in the adjustment amount on the reward; Indicates the fine-tuning coefficient (usually 0.01).

[0079] The final concentration of hydrogen-rich water is:

[0080] Where: Indicates the correction factor (i.e., the correction factor based on the physiological characteristics of the crop at the current growth stage); when the plant is in the flowering stage, When the plant is in the seedling stage, .

[0081] Example 2: As a further optimization of the present application, based on the solution of Example 1, the chlorophyll fluorescence parameters include the maximum optical efficiency ( F v / F m , reflecting the potential activity of the photosystem II reaction center. Photosystem II is an important protein complex located on the thylakoid membrane in photosynthesis. Its main function is to absorb light energy and catalyze the photolysis of water to produce oxygen, protons and electrons) and the photochemical quenching coefficient (qP, reflecting the open ratio of the photosystem II reaction center):

[0082] Where: F 0 represents the initial fluorescence intensity; F m represents the maximum fluorescence intensity; F v represents variable fluorescence intensity; F t represents the steady-state fluorescence intensity under light adaptation; represents the maximum fluorescence intensity under light adaptation; represents the initial fluorescence intensity under light adaptation; Canopy multispectral parameters include the Normalized Difference Vegetation Index (NDVI) and the Red Edge Position (REP, a sensitive indicator reflecting changes in chlorophyll content):

[0083] Where: represents the reflectivity in the near-infrared band; Indicates the reflectivity of red light band; 、 、 Respectively represent the reflectivity of 680nm, 705nm, and 750nm bands; Micro-deformation of the stem (measured by strain sensor):

[0084] Where: Indicates the change in stem length; represents the initial length of the stem; Leaf infrared temperature (canopy temperature, measured by infrared sensor):

[0085] Where: C1, C2 represent the Planck constant related parameters (C1 is , C2 is ); Indicates the spectral radiance received by the sensor; Indicates the wavelength of infrared band; Root growth parameters (obtained through 3D image scanning) include root length density RLD (reflecting the root absorption range) and root surface area index RSI (reflecting absorption capacity):

[0086] Where: L 根 Indicates the root length; V 土壤 represents the corresponding soil volume; r 根 represents the root radius.

[0087] Field environmental factor data, such as light intensity, air temperature and humidity, soil temperature and humidity, CO2 concentration, and soil pH value, can be obtained through corresponding light intensity sensors, air temperature and humidity sensors, soil temperature and humidity sensors, CO2 concentration sensors, and soil pH value sensors.

[0088] Example 3: As a further optimization of the solution of the present application, a hydrogen-rich water concentration dynamic adjustment system is also included, which adopts any method in Example 1 or Example 2 to achieve hydrogen-rich water concentration adjustment during crop planting.

Claims

1. A method for dynamically adjusting the concentration of hydrogen-rich water, characterized by: include: Step S1, data collection: deploying data collection sensors based on crop types and field conditions to collect plant growth trend data and field environmental factor data respectively; Step S2, data preprocessing: performing noise removal, abnormal data detection and correction, and data normalization preprocessing on the collected data; Step S3, extraction of key influencing factors: exploring plant growth trend data and field environmental factor data to determine the key factors affecting crop growth; Step S4, obtaining the initial concentration of hydrogen-rich water: inputting the key influencing factor data into the fuzzy logic system, defining the fuzzy subsets of each key influencing factor, and obtaining the initial concentration of hydrogen-rich water; Step S5, obtaining the optimized concentration of hydrogen-rich water: obtaining the dynamic concentration adjustment amount, extreme environment adjustment amount, and implicit stress adjustment amount through reinforcement learning dynamic optimization, extreme environment adaptation, and implicit stress early warning, respectively, to obtain the final concentration of hydrogen-rich water.

2. A method for dynamically adjusting the concentration of hydrogen-rich water according to claim 1, characterized in that: The growth trend data include chlorophyll fluorescence, canopy multispectral, stem micro-deformation degree, leaf infrared temperature, and root growth; the field environmental factor data include light intensity, air temperature and humidity, soil temperature and humidity, CO2 concentration, and soil pH value.

3. A method for dynamically adjusting the concentration of hydrogen-rich water according to claim 1, characterized in that: The noise removal in step S2 is to remove random noise using a Kalman filter algorithm.

4. A method for dynamically adjusting the concentration of hydrogen-rich water according to claim 1, characterized in that: The abnormal data detection and correction in step S2 are specifically as follows: For input feature data X =[ x 1, x 2,…, x n ], the isolation forest algorithm is used to detect anomalies in the data after random noise is removed, and the data anomaly score is obtained: Where: h(X) Represents a sample X path length in the isolation tree; E[h(X)] represents the expected path length; represents Euler's constant; Preset anomaly score threshold SX d ,like , then the corresponding sample is determined to be an outlier; For outliers, Huber robust regression is used for correction, and the Huber loss function is: Where: is a constant.

5. A method for dynamically adjusting the concentration of hydrogen-rich water according to claim 1, characterized in that: The step S3 of extracting key influencing factors specifically comprises: extracting key factors based on key point identification of the "stress-metabolism" coupling network, correlation analysis of dynamic time lag, key variable allocation of multi-scale feature fusion, and dynamic key factors adaptive to stress state. If a variable is identified as a key influencing factor by at least three of the four methods, namely, key point identification of the "stress-metabolism" coupling network, correlation analysis of dynamic time lag, key variable allocation of multi-scale feature fusion, and dynamic key factors adaptive to stress state, then the variable is a key influencing factor. If a variable is identified as a key influencing factor by any of the four methods: key point identification based on the "stress-metabolism" coupling network, dynamic time-lag correlation analysis, key variable allocation based on multi-scale feature fusion, and stress state adaptive dynamic key factors, it will be directly excluded; If a variable is identified as a key influencing factor by any two of the four methods: key point identification based on the "stress-metabolism" coupling network, correlation analysis of dynamic time lag, key variable allocation based on multi-scale feature fusion, and dynamic key factors based on stress state adaptation, it will be regarded as a secondary influencing factor.

6. A method for dynamically adjusting the concentration of hydrogen-rich water according to claim 5, characterized in that: Identification of key points based on the "stress-metabolism" coupling network: Establishment of a reactive oxygen species level estimation model: Where: T leaf Indicates the infrared temperature of the leaves, T air Indicates air temperature; represents the volumetric moisture content of the soil, represents field capacity; F v / F m represents the chlorophyll fluorescence parameter; K 1. K 2. K 3 represents the corresponding weight coefficient; represents the error term; Based on the ROS level estimation model and the partial least squares path model, the preprocessed data variables were divided into three layers: the environmental layer of the ROS generation driving source, the physiological layer of the ROS damage target, and the effect layer of the ROS regulation results. Based on the mediating role of ROS, the path CeX→CeY and the path CeY→CeZ were established, and the path coefficients of the two paths were obtained: 、 ; Preset path coefficient threshold ,like , then the corresponding variables are taken as key influencing factors; Dynamic time-lag correlation analysis: Quantifying the dynamic correlation between variables using time-lag mutual information: Growth trends or field environmental factor variables X(t) and concentration requirements , in the time lag Mutual information under TDMI for: Where: P represents the probability distribution, This means that the variable X is at time t Pick a Value and concentration demand B at time Pick b The probability of the value; Calculate the TDMI hysteresis of each growth trend or field environmental factor variable separately. The variable corresponding to the peak value of TDMI is taken as the key influencing factor; Key variable allocation for multi-scale feature fusion: For time series variables X(t) Perform three-layer wavelet packet decomposition to obtain the energy values ​​of eight frequency bands E i,j : Where: i Indicates the variable number; j Indicates the frequency band number; represents the wavelet basis function; The sub-signal of each frequency band is input into the random forest as an independent feature to obtain the Gini importance of each sub-signal; based on the metabolic cycle of hydrogen-rich water, different frequency bands are given weights. , and obtain the comprehensive importance of the corresponding variables: Where: Gini i,j Indicates the i The variable j Gini importance of frequency band features; Preset comprehensive importance threshold I d ,If the comprehensive importance of a variable is greater than the comprehensive importance threshold, the corresponding variable is regarded as a key influencing factor; Key factors for adaptive dynamics of stress states: Introduction of stress index SI: Where: T opt Indicates the most suitable leaf temperature; T max Indicates critical high temperature; K 4. K 5 respectively represent the corresponding weight coefficients; Preset stress index threshold SI d : when When , it means that the plant metabolism is basically normal. The posterior probability of the variables and the concentration of hydrogen-rich water is calculated through the Bayesian network. The variables are arranged from large to small and the top three variables are taken as key influencing factors. when When , it means that the photosynthesis / respiration of the plant is inhibited. The posterior probability of the variables and the concentration of hydrogen-rich water is recalculated through the Bayesian network, and the variables are arranged from large to small and the top three variables are taken as key influencing factors.

7. A method for dynamically adjusting the concentration of hydrogen-rich water according to claim 6, characterized in that: The step S4 is specifically as follows: Define the membership function of the fuzzy subset of key influencing factors, including a small number of membership U dr (S) , appropriate membership U su (S) Excess membership U we (S) : Where: S represents the current measured value of the key influencing factor; S min Indicates the critical value when the key influencing factor is too low; S a Indicates the appropriate central value of the key influencing factors; S max Indicates the critical value when the key influencing factor is too high; The fuzzy rule base is constructed by integrating multi-factor association and expert practical experience, and the center of gravity method is used to defuzzify and obtain the preliminary concentration: Where: U i Indicates the i The membership degree of the rule; C i Indicates the i Reference concentration for the rule; m Indicates the number of activated rules.

Images