Sweet potato water culture seedling intelligent water and fertilizer management method and system

Abstract

The application discloses an intelligent water and fertilizer management method and system for sweet potato water culture seedlings, which analyzes nutrient deficiency by water culture tank in-situ monitoring data flow, and abstracts global water and fertilizer target concentration scalar; then, a liquid flow convection and diffusion partial differential equation model is constructed according to the geometric size of the tank body and the flow velocity of the main circulating pump, the scalar is taken as a first end excitation for numerical inversion, and the lag time of concentration shock reaching each place along the way is obtained, that is, a space phase shift matrix; then, the valve coordinates are taken as indexes, the lag along the way is decoupled into liquid injection advance and pulse volume instructions of each node, and single global decision is disassembled into time and space synchronous distributed actions; finally, the micro-fluidic valves along the tank are triggered according to the advance amount, and the pulse width is corrected through monitoring data feedback, so that the out-of-phase injection of mother liquor is superimposed along the way in the low flow liquid, the concentration of the whole tank is synchronized to reach the standard without improving the flow rate, and the seedling burning at the first end and the nutrient deficiency at the tail end are eliminated.

Description

Technical Field

[0001] This application relates to the field of intelligent agricultural water and fertilizer regulation technology, and more specifically, to an intelligent water and fertilizer management method and system for hydroponic sweet potato seedlings. Background Technology

[0002] Sweet potato, as an important dual-purpose crop for both food and forage, commonly employs hydroponics for the large-scale rapid propagation of virus-free seedlings. Hydroponic seedlings lack the nutrient buffering and ion adsorption capacity of soil substrates, relying entirely on circulating nutrient solutions for nitrogen, phosphorus, and potassium. However, the single-batch seedling cultivation cycle for rapid propagation is typically only ten days, leading to drastic fluctuations in nutrient supply and demand. Excessive fertilization results in excessive vegetative growth without root development, while insufficient fertilization leads to delayed adventitious root development and weak stems, directly impacting transplant survival rates. Furthermore, industrial-scale seedling troughs often reach tens or even hundreds of meters in length, making manual sampling and measurement insufficient to monitor nutrient status across all sections. Therefore, it is necessary to develop an intelligent water and fertilizer management method and system for sweet potato hydroponics to achieve automatic sensing and precise allocation of nutrient solution concentration.

[0003] Existing intelligent water and fertilizer management methods and systems for sweet potato hydroponics mostly involve deploying conductivity, pH, and ion concentration sensors in the storage tank, combining them with a pre-set crop growth model or reinforcement learning algorithm to calculate the globally optimal water and fertilizer ratio, and then injecting the prepared nutrient solution into the tank for circulation and renewal via a mother liquor pump and a main circulation pump. This approach treats the entire long-distance hydroponic tank as a lumped parameter system, outputting only a single global action and implicitly assuming that the action takes effect immediately. However, to avoid tearing the fragile primary adventitious roots of the sweet potato seedlings by the shear force of the water flow, the circulation velocity within the tank is strictly limited to an extremely low range. The nutrient solution concentration wave takes tens of minutes or even hours to transport from the beginning to the end via advection and diffusion, creating a huge pure lag time. Using a lumped action to control a time-delayed object that is essentially a distributed parameter system means that different sections are in different concentration phases for extended periods. Seedlings at the beginning are susceptible to ion poisoning and burning due to instantaneous high concentrations, while seedlings at the end remain in a state of latent nutrient deficiency. This makes it difficult to guarantee batch uniformity and seedling vigor in large-scale rapid propagation. Furthermore, even if the liquid is injected in a dispersed manner along the tank, multiple injection valves share the same mother liquor main pipe. When their operating windows overlap on the time axis, it will cause transient pressure drop and flow rate decay, resulting in a systematic underrepresentation of the actual injected mother liquor volume at each node, which in turn forms a persistent concentration depression zone in the middle of the tank.

[0004] Therefore, an optimized intelligent water and fertilizer management solution for sweet potato hydroponic seedlings is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an intelligent water and fertilizer management method and system for hydroponic sweet potato seedlings.

[0006] According to one aspect of this application, an intelligent water and fertilizer management method for hydroponic sweet potato seedlings is provided, comprising: S1. Based on the nutrient demand model of sweet potato hydroponic seedlings, nutrient deficiency analysis is performed on the collected in-situ monitoring data stream of hydroponic troughs to obtain the global target concentration of water and fertilizer. S2, based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic tank, the geometric dimension vector of the long-distance hydroponic tank is discretized in one dimension and mapped by advection dynamics to construct a partial differential equation model of liquid flow convection diffusion. S3 uses the global target concentration of water and fertilizer as the scalar value of the first-end excitation injection liquid flow convection diffusion partial differential equation model, and performs numerical inversion on the propagation delay of the nutrient solution concentration shock wave along the path to obtain the spatiotemporal phase shift characteristic matrix. S4. Based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank, the physical space discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix and the decoupling of the distributed control command are performed to obtain the distributed injection advance time series and the microfluidic pulse volume command of each node. S5 is a microfluidic valve uniformly distributed along the tank based on the phase-controlled trigger of the distributed injection advance time series. The injection volume of the mother liquor is controlled by the microfluidic pulse volume command of each node, and the pulse width is corrected by the feedback of the in-situ monitoring data stream of the hydroponic tank so that the heterogeneous injected mother liquor is superimposed along the flow in the low-flow-rate circulating liquid flow.

[0007] According to another aspect of this application, an intelligent water and fertilizer management system for hydroponic sweet potato seedlings is provided, comprising: The nutrient deficiency analysis module is used to analyze the nutrient deficiency of the collected in-situ monitoring data stream of the hydroponic trough based on the nutrient demand model of sweet potato hydroponic seedlings in order to obtain the global target concentration of water and fertilizer. The fluid transport modeling module is used to perform one-dimensional spatial discretization and advection dynamic mapping on the geometric dimension vector of a long-distance hydroponic tank based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic tank, so as to construct a partial differential equation model of fluid convection and diffusion. The spatiotemporal phase shift inversion module is used to take the global water and fertilizer target concentration scalar as the first-end excitation injection liquid flow convection diffusion partial differential equation model, and to perform numerical inversion on the propagation delay of the nutrient solution concentration shock wave to obtain the spatiotemporal phase shift characteristic matrix. The distributed command decoupling module is used to perform physical space discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix and decouple it from the distributed control command based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank, so as to obtain the distributed injection advance time series and the microfluidic pulse volume command of each node. The phase-controlled liquid injection execution module is used to trigger microfluidic valves evenly distributed along the tank based on the phase-controlled time series of distributed liquid injection advance. The injection volume of mother liquor is controlled by the microfluidic pulse volume command of each node, and the pulse width is corrected by the feedback of the in-situ monitoring data stream of the hydroponic tank so that the heterogeneous injected mother liquor is superimposed along the flow in the low-flow-rate circulating liquid flow.

[0008] Compared with existing technologies, this invention proposes an intelligent water and fertilizer management method for sweet potato hydroponic seedlings. First, the nutrient deficit level during the current growth stage is analyzed from the in-situ monitoring data stream of the hydroponic tank, condensed into a global target water and fertilizer concentration scalar, clarifying the concentration that the entire tank should achieve. Then, a partial differential equation model of liquid flow convection and diffusion is established based on the tank's geometry and the main circulation pump's flow rate. Using this target concentration as the initial excitation, numerical inversion is performed to calculate the hysteresis time required for the concentration shock wave to reach various points along the flow path, i.e., the spatiotemporal phase shift characteristic matrix, thus transforming the invisible transport delay into a calculable phase quantity. Furthermore, using valve coordinates as an index, the hysteresis time along the flow path is decoupled into the injection advance amount and pulse volume command for each node, breaking down the originally singular global decision into a set of spatiotemporally synchronized distributed actions. Finally, the microfluidic valves evenly distributed along the tank are triggered in advance and the pulse width is corrected by the feedback of monitoring data. This allows the mother liquor injected in different phases to be superimposed along the flow in the low-flow-rate circulating liquid. Without increasing the flow rate or changing the root-nourishing microenvironment, the concentration of the entire tank is simultaneously achieved, eliminating the mismatch between the burning of seedlings at the beginning and the lack of nutrients at the end. Attached Figure Description

[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flowchart of an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application; Figure 2 This is a data flow diagram of an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application; Figure 3 The flowchart illustrates a method for intelligent water and fertilizer management of sweet potato hydroponic seedlings according to an embodiment of this application, which uses the global target concentration of water and fertilizer as the initial excitation injection liquid flow convection diffusion partial differential equation model to numerically invert the propagation delay of the nutrient solution concentration shock wave to obtain the spatiotemporal phase shift characteristic matrix. Figure 4Based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic trough, the valve coordinates are determined by the valve coordinates. The physical space discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix is ​​performed and the distributed control command is decoupled to obtain the distributed injection advance time series and the flowchart of the microfluidic pulse volume command of each node. Figure 5 This is a flowchart illustrating an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application, which uses distributed injection advance time series phase-controlled triggering microfluidic valves evenly distributed along the tank. The flowchart describes a method that controls the injection volume of mother liquor by microfluidic pulse volume commands at each node and corrects the pulse width by feedback from in-situ monitoring data streams in the hydroponic tank, so that the heterogeneous injected mother liquor is superimposed along the flow path in the low-flow-rate circulating liquid flow. Figure 6 This is a flowchart illustrating an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application, which uses adaptive pulse width calculation based on the transient voltage drop coupling matrix between nodes to obtain a multi-node transient physical drive instruction set. Figure 7 This is a block diagram of an intelligent water and fertilizer management system for sweet potato hydroponic seedlings according to an embodiment of this application. Detailed Implementation

[0011] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0012] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0013] While this application makes various references to certain modules of the system according to embodiments of this application, these modules can be deployed in a seedling nursery water and fertilizer management server, a hydroponic trough field edge controller, and / or a programmable logic controller. Modules can be combined or split depending on computing resources. The modules are illustrative only, and different aspects of the system and method may use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0015] Figure 1 This is a flowchart of an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application. Figure 1 and Figure 2 As shown, an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application includes the following steps: S1, based on the nutrient demand model of sweet potato hydroponic seedlings, performing nutrient deficiency analysis on the collected in-situ monitoring data stream of the hydroponic trough to obtain a global water and fertilizer target concentration scalar; S2, based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic trough, performing one-dimensional spatial discretization and advection dynamics mapping on the geometric dimension vector of the long-distance hydroponic trough to construct a liquid flow convection diffusion partial differential equation model; S3, using the global water and fertilizer target concentration scalar as the initial excitation injection liquid flow convection diffusion partial differential equation model, and controlling the nutrient solution concentration shock wave along the... S4. Numerical inversion of the propagation delay is performed to obtain the spatiotemporal phase shift feature matrix; S5. Based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank, the physical space discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix is ​​performed and the distributed control command is decoupled to obtain the distributed injection advance time series and the microfluidic pulse volume command of each node; S6. Based on the distributed injection advance time series, the microfluidic valves uniformly distributed along the tank are phase-controlled triggered, wherein the injection volume of mother liquor is controlled by the microfluidic pulse volume command of each node and the pulse width is corrected by the feedback of the in-situ monitoring data stream of the hydroponic tank so that the heterogeneous injected mother liquor is superimposed along the path in the low-flow-rate circulating liquid flow.

[0016] Specifically, in step S1, based on the sweet potato hydroponic seedling nutrient requirement model, nutrient deficiency analysis is performed on the collected in-situ monitoring data stream from the hydroponic trough to obtain a global target concentration scalar value for water and fertilizer. It should be noted that, because the nitrogen, phosphorus, and potassium requirements of sweet potato hydroponic seedlings differ significantly between the rooting and vigorous growth stages, the total ion concentration reflected by a single conductivity value cannot determine which elemental dimension is deficient. Therefore, the technical solution of this application first performs nutrient deficiency analysis on the collected in-situ monitoring data stream from the hydroponic trough based on the sweet potato hydroponic seedling nutrient requirement model to obtain a global target concentration scalar value for water and fertilizer. Through the above processing, a unique and clear initial concentration excitation benchmark can be provided for the subsequent numerical inversion of propagation delay along the path.

[0017] More specifically, in a specific example of this application, step S1 includes: based on the agronomic nutritional knowledge graph of sweet potato hydroponic seedlings, mapping the growth stage state and calibrating the ion concentration baseline of the in-situ monitoring data stream of the hydroponic trough to obtain a standard nutrient requirement concentration vector and baseline ion concentration values ​​at the beginning and end; performing dynamic nutrient deficit calculation on the standard nutrient requirement concentration vector and baseline ion concentration values ​​at the beginning and end to obtain the system nutrient deficit characteristics; and performing state value optimization and action decision-making on the system nutrient deficit characteristics to obtain a global water and fertilizer target concentration scalar. The sweet potato hydroponic seedling nutrient requirement model consists of two parts: a sweet potato hydroponic seedling agronomic nutritional knowledge graph and a state value decision-making network. The former is responsible for mapping from the growth stage to the suitable concentration range, and the latter is responsible for decision-making from the deficit to the target concentration.

[0018] More specifically, ion-selective electrodes, conductivity electrodes, and liquid temperature probes, positioned at the beginning and end of the long-distance hydroponic tank, synchronously sample at fixed intervals. The collected ion concentration, conductivity, and liquid temperature signals from the beginning and end of the nutrient solution, along with the main circulation pump flow rate parameters, are combined with the sweet potato hydroponic seedling growth stage category tags maintained by the host computer to form an in-situ monitoring data stream for the hydroponic tank. Using a sweet potato hydroponic seedling agronomic nutrition knowledge graph as the mapping basis, this graph uses growth stages such as rooting and vigorous vine development as nodes, and the appropriate concentration range of each core nutrient element at the corresponding stage as node attributes. The growth stage category tags are searched and matched in the graph to read the appropriate concentration values ​​for the current stage. For example, for the rooting stage, nitrate nitrogen is 120 mg / L, phosphorus is 30 mg / L, and potassium is 150 mg / L. These values ​​are then serialized into a standard nutrient requirement concentration vector according to a fixed element order. Simultaneously, baseline calibration was performed on the ion concentration signals at the beginning and end of the sampling process. First, the liquid temperature signal was read, with 25 degrees Celsius as the reference temperature. The conductivity electrode was deviated from this reference temperature by approximately 2% for every 1 degree Celsius deviation. The temperature coefficient of the ion-selective electrode was determined according to the relationship between the slope and absolute temperature in the Nernst equation. The conductivity and ion concentration readings were then converted point-by-point to the 25-degree Celsius reference temperature to eliminate electrode output drift caused by the diurnal temperature difference in the greenhouse nutrient solution. Subsequently, the converted ion concentration components were weighted and summed according to their equivalent concentrations. This sum was compared with the total ion concentration converted from the conductivity signal at the same sampling time. If the relative deviation was within 10%, the set of readings was considered self-consistent. If it exceeded this range, the corresponding electrode surface was considered to have a biofilm formed by sweet potato root exudates, and the set of readings was discarded. The arithmetic mean of the effective readings at the beginning and end that passed the order of magnitude verification was taken to obtain the reference ion concentration values ​​for both ends. The flow rate parameter of the main circulation pump is not involved in the concentration calculation of the nutrient dimension in this step. Its non-zero state is used to confirm that the circulating liquid flow is in normal transport conditions, thereby ensuring that the readings at the beginning and end are averaging. It is also used as bypass data to be transmitted through the in-situ monitoring data flow of the hydroponic tank, and is used as the source of the advection term when constructing the liquid flow convection diffusion partial differential equation model later.

[0019] Based on the standard nutrient requirement concentration vector and the baseline ion concentration values ​​at both ends, the standard nutrient requirement concentration vector and the baseline ion concentration values ​​at both ends are aligned one by one according to the ion dimension. For each dimension, the difference between the standard requirement value and the measured baseline value is calculated. The rectification constraint ensures that the dimension participates in the accumulation only when the measured value is lower than the standard requirement value. Then, the difference of each dimension is squared and multiplied by the sensitivity weighting coefficient of the corresponding ion. After accumulation, the square root is taken to obtain the system nutrient deficit characteristic in the form of weighted L2, as follows: ; in, This represents the characteristics of the system's nutrient deficit. This represents the total number of core nutrient ions included in the monitoring. This refers to the ion dimension number. For the first The sensitivity weighting coefficients for seed ions were set to 0.2, 0.5, and 0.3 respectively during the rooting stage, according to the fixed dimensional order of nitrate nitrogen, phosphorus, and potassium. The value is set to 3, which includes three core nutrient ions: nitrate nitrogen, phosphorus, and potassium. Among them, phosphorus has the highest sensitivity weighting coefficient to match the phosphorus dependence of adventitious root formation at this stage. The first in the standard nutrient requirement concentration vector The appropriate concentration value of seed ions, The first and last reference ion concentration values ​​are the first and last reference ion concentration values. Measured values ​​of seed ions, This is the positive deficit component of this dimension. When the measured concentration of a certain ion is higher than the appropriate concentration, this component is set to 0, thereby avoiding the need to continue supplementing the already excessive potassium, which would exacerbate the antagonism between potassium and calcium and cause ion poisoning in seedlings.

[0020] Based on this, the system's nutrient deficit characteristics are defined as the environmental observation state at the current moment in the Markov decision-making process. The range of injectable nutrient solution concentrations at the beginning is discretized into a set of candidate actions with a step size of 10 mg / L. The immediate reward function uses the convergence rate of the deficit as the positive term and the amount exceeding the upper limit of the safe concentration for sweet potato hydroponic seedlings and the amount of mother liquor consumed in this step as the penalty term. The discount factor is set to 0.9 to balance the current fertilization response and the concentration stability of the entire rapid propagation batch. Based on this, the state value of each candidate action is optimized according to the Bellman optimal equation, and the action that maximizes the cumulative expected return is taken as the decision result, as shown below: ; in, This serves as a scalar for the overall target concentration of water and fertilizer. For candidate injection concentration action, To serve as a characteristic of the system's nutrient deficit in the current environmental observation state, To perform actions under the current nutrient deficiency state The instant rewards received As a discount factor, To perform the action The next time step after the hydroponic tank evolves into a deficit state. To sum over all possible deficit states in the next time step, To perform an action in the current deficit state. Then transferred to state The state transition probability, Let be the optimal state value function for the next state. To select the candidate injection concentration action that maximizes the value within the square brackets, the combination term within the square brackets represents the action. The comprehensive evaluation value is based on the current agricultural benefits and the potential benefits of the subsequent seedling cycle. The global water and fertilizer target concentration scalar calculated based on this is a fixed concentration value, such as 120 mg / L. Since the nitrogen, phosphorus, and potassium elements in the mother liquor tank are pre-mixed according to the proportion determined by the standard nutrient requirement concentration vector of the current growth period, the mother liquor ratio remains constant during a single injection. Therefore, the concentration of each element changes synchronously with this scalar in a fixed proportion. This scalar, characterized by the nitrate nitrogen concentration, can uniquely determine the multi-element formula of the nutrient solution. This value serves as the concentration setting benchmark at the first injection point for use in subsequent stages.

[0021] Specifically, the reinforcement learning model used in the aforementioned state value optimization and action decision-making processes employs a deep Q-network algorithm. Its value network consists of a three-layer fully connected structure. The input layer receives a state vector composed of system nutrient deficit characteristics, sweet potato hydroponic seedling growth stage category labels, and liquid temperature signals. Each of the two hidden layers contains sixty-four neurons, using modified linear units as activation functions. The number of neurons in the output layer is the same as the number of discretized candidate injection concentration actions. Each output value represents the state-action value estimate for the corresponding action. The state transition probability is not explicitly modeled but implicitly provided in each interaction step by a simulation environment jointly constructed by the sweet potato hydroponic seedling nutrient absorption rate model and the liquid flow convection-diffusion partial differential equation model. That is, the current concentration and absorption rate are substituted into the model to extrapolate to the next sampling period to return to the subsequent state. The immediate reward function is expressed as: ; in, To perform an action in the current deficit state. The instant reward received This represents the decrease in the nutrient deficit characteristics of the system in this step. This refers to the candidate injection concentration action performed in this step. The upper limit of safe concentration for sweet potato hydroponic seedlings, This value represents the amount by which the action exceeds the upper limit of the safe concentration. The value is zero if the action value does not exceed this upper limit. This refers to the total amount of mother liquor injected into the entire tank in this step. to The three weights represent the positive benefit of deficit convergence, the penalty for ion toxicity caused by concentration exceeding limits, and the cost of mother liquor consumption, respectively. Each item is normalized according to its range before weighting, so each weight coefficient is dimensionless. The training data is taken from the in-situ monitoring records of rapid propagation batches of virus-free sweet potato seedlings over the years, including the ion concentration, conductivity, and liquid temperature time series of each growth stage, as well as the number of adventitious roots and transplant survival rate of the corresponding batches. After normalization, the data, along with the interactive samples generated by the simulation environment, are stored in the experience replay pool. During training, candidate actions are explored randomly with a fixed probability. Small batches of samples are randomly selected from the replay pool, and the value network parameters are updated using the mean square value of the time difference error as the loss function. The parameters are copied to the target network at fixed intervals to stabilize the value estimation. Once the loss converges and the number of concentration exceeding limits of the verification batch is lower than the set value, the parameters are solidified for online decision-making.

[0022] Specifically, in step S2, based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic trough, the geometric dimension vector of the long-distance hydroponic trough is discretized in one dimension and mapped using advection dynamics to construct a partial differential equation model for liquid flow convection and diffusion. It should be noted that because the primary adventitious roots of sweet potato hydroponic seedlings are intolerant of water flow shear, the circulation velocity within the trough is limited to a low range, and the transport lag of nutrient solution concentration along the longitudinal direction of the trough cannot be estimated empirically. Therefore, the technical solution of this application further uses the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic trough to discretize the geometric dimension vector of the long-distance hydroponic trough in one dimension and map using advection dynamics to construct a partial differential equation model for liquid flow convection and diffusion. Through the above processing, the concentration transport process within the trough can be transformed into a numerically solvable distributed parameter object.

[0023] More specifically, in a specific example of this application, step S2 includes: calculating the cross-sectional area and converting the cross-sectional velocity of the long-distance hydroponic trough geometric dimension vector and the main circulation pump velocity parameters to obtain a one-dimensional advection characteristic velocity scalar and the effective flow cross-sectional area of ​​the hydroponic trough; converting the long-distance hydroponic trough geometric dimension vector and the effective flow cross-sectional area of ​​the hydroponic trough into advection dynamic parameters to obtain a one-dimensional longitudinal spatial domain vector; and assembling the one-dimensional longitudinal spatial domain vector and the one-dimensional advection characteristic velocity scalar into a convection-diffusion operator to construct a liquid flow convection-diffusion partial differential equation model.

[0024] In other words, more specifically, the first step is to read the geometric dimension vector of the long-distance hydroponic trough. This vector consists of three parameters: the total length of the trough along its length, the internal bottom width, and the effective liquid level depth. In a seedling trough used for rapid propagation of virus-free sweet potato seedlings, these are 60m, 0.25m, and 0.04m, respectively. Since the roots of the sweet potato hydroponic seedlings are suspended in the liquid layer and the liquid layer thickness is much smaller than the trough length, the lateral and vertical concentration gradients can be ignored. Therefore, the internal bottom width and the effective liquid level depth are multiplied to calculate the effective flow cross-sectional area of ​​the hydroponic trough perpendicular to the liquid flow direction, as shown below: ; in, The effective cross-sectional area for circulation in the hydroponic tank. The width of the bottom of the hydroponic tank is the internal width of the tank, which is obtained by resolving the geometric dimension vector of the long-distance hydroponic tank. The effective liquid level depth for operation, calculated based on the above values, yields an effective flow cross-sectional area of ​​0.01 m² for the hydroponic tank. Subsequently, the main circulation pump velocity parameter is extracted from the in-situ monitoring data stream of the hydroponic tank. This parameter represents the volumetric flow rate injected into the tank by the main circulation pump at the current speed, and is taken as 0.6 m³ / h, or 1.67 × 10⁻⁶ m³ / h. -4 m³ / s; Dividing this volumetric flow rate by the effective cross-sectional area of ​​the hydroponic tank converts the hardware flow parameters of the pumping process into the linear velocity of the liquid particles advancing along the longitudinal axis of the tank, yielding a one-dimensional advection characteristic velocity scalar, expressed as follows: ; in, For a one-dimensional advection characteristic velocity scalar, The main fluid volumetric flow rate is characterized by the flow velocity parameters of the main circulation pump. Given the effective flow cross-sectional area of ​​the hydroponic trough, the one-dimensional advection characteristic velocity scalar calculated using the above values ​​is 1.67 × 10⁻⁶. - The flow rate of 2 m / s is within the shear tolerance range of the primary adventitious roots of sweet potato seedlings, and it takes about 3600 seconds for the concentration front to travel from the beginning to the end at 60m.

[0025] Based on the effective flow cross-sectional area of ​​the hydroponic trough, the total length of the trough along the geometric dimension vector of the long-distance hydroponic trough is taken as the upper bound of the spatial domain. Starting from the zero coordinate of the liquid inlet at the first end, points are equidistantly arranged along the longitudinal axis to discretize the continuous spatial domain into a finite number of nodes. The discrete step length is determined based on the spacing of the microfluidic valves to be deployed along the trough, and is taken as an integer fraction of the valve spacing, so that the coordinates of each valve must fall on the coordinates of the discrete nodes. At the same time, the liquid volume of a single discrete unit is on the same order of magnitude as the local liquid volume under the jurisdiction of a single valve. In the above trough, the valve spacing is 1m, the discrete step length is 0.5m, the volume of a single discrete unit is 5L and the volume of a single control cluster is 10L. Based on this, a one-dimensional longitudinal spatial domain vector containing 121 nodes is obtained, and the vector elements are the longitudinal coordinate values ​​of each node. Meanwhile, the time step size is constrained by the ratio of the offset step size to the one-dimensional advection characteristic velocity scalar, so that the Coulomb number is no greater than one. Under the above values, this ratio is about 30s, and the upper limit of diffusion stability determined by the ratio of the square of the offset step size to twice the longitudinal hydrodynamic dispersion coefficient is about 104s. Therefore, the time step size is taken as 5s to ensure that the displacement of the concentration front in a single time step does not exceed one offset step size in the subsequent numerical solution process.

[0026] Based on this, a convection-diffusion operator was assembled between the one-dimensional longitudinal spatial domain vector and the one-dimensional advection characteristic velocity scalar: a longitudinal hydrodynamic dispersion coefficient was introduced to characterize the combined effect of molecular diffusion and mechanical dispersion caused by the sweet potato root network. This coefficient was pre-calibrated by tracer impulse response experiments, and a value of 1.2 × 10⁻⁶ was used in the aforementioned tank. - ³m² / s; Using coordinate operators provided by a one-dimensional longitudinal spatial domain vector, construct the first and second partial derivatives of concentration with respect to spatial coordinates. Multiply the second partial derivative by the dispersion coefficient to form the diffusion term, and multiply the first partial derivative by the one-dimensional advection characteristic velocity scalar to form the convection term. Combine these with the partial derivative of concentration with respect to time to construct the partial differential equation model of liquid flow convection and diffusion, as follows: ; in, Vertical axis At the moment The transient ion concentration of the nutrient solution. These are the vertical spatial coordinates extracted from the one-dimensional vertical spatial domain vector. For time, The longitudinal hydrodynamic dispersion coefficient, For a one-dimensional advection characteristic velocity scalar, Let be the rate of change of concentration at a fixed point in space over time. The diffusion term characterizes the effect of gradual smoothing of the interface with abrupt concentration changes under the influence of dispersion. The convection term characterizes the effect of the overall concentration distribution shifting towards the tail end with the liquid flow. The sum of the two terms constitutes a complete dynamic description of the evolution of nutrient solution concentration along the tank under low flow rate circulation conditions, which can be used for subsequent numerical inversion with the global water and fertilizer target concentration scalar as the initial excitation.

[0027] Specifically, in step S3, the global target water and fertilizer concentration scalar is used as the partial differential equation model of the convection and diffusion of the injected liquid at the first end. The propagation delay of the nutrient solution concentration shock wave along the path is numerically inverted to obtain the spatiotemporal phase shift characteristic matrix. It should be noted that, given that the liquid convection and diffusion partial differential equation model only describes the general law of concentration evolution along the tank and has not yet established a correspondence with the nutrient target required by the current batch of sweet potato seedlings, and that under low-flow-rate circulation conditions, the concentration front takes a long time to travel from the first end to the last end, and the diffusion effect causes the concentration interface to gradually flatten along the path, the actual time at which the target concentration is reached at each valve coordinate cannot be directly estimated by dividing the flow rate by the distance. If this difference along the path is not pre-quantified, the injection actions can only be performed simultaneously at the same time, and the sweet potato seedlings at the last end will continue to be in a low-concentration environment. Based on this, the technical solution of this application further uses the global target water and fertilizer concentration scalar as the partial differential equation model of the convection and diffusion of the injected liquid at the first end to numerically invert the propagation delay of the nutrient solution concentration shock wave along the path to obtain the spatiotemporal phase shift characteristic matrix. Through the above processing, the transport hysteresis that cannot be directly observed can be transformed into a calculable phase quantity that corresponds one-to-one with each longitudinal coordinate, providing a quantitative basis for decoupling a single global concentration decision into a distributed injection action.

[0028] Figure 3 This document describes a flowchart illustrating an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application. It describes a partial differential equation model using the global target water and fertilizer concentration as the initial excitation of the injected liquid flow convection and diffusion, and a numerical inversion of the propagation delay of the nutrient solution concentration shock wave to obtain the spatiotemporal phase shift characteristic matrix. Figure 3 As shown, step S3 includes: S31, setting the global water and fertilizer target concentration scalar as the first type of boundary condition at the beginning, and performing boundary condition constraints and equation closure processing on the liquid flow convection diffusion partial differential equation model to obtain the liquid flow boundary condition differential equation set; S32, solving the liquid flow boundary condition differential equation set by solving the concentration shock wave propagation trajectory along the path to obtain the full-time-space concentration response function; S33, according to the preset steady-state concentration tolerance threshold coefficient, performing node-by-node target attainment time underfimum inverse solution on the full-time-space concentration response function to generate the spatiotemporal phase shift characteristic matrix.

[0029] In step S31, the global target water and fertilizer concentration scalar is set as the first type of boundary condition at the beginning. The partial differential equation model of liquid flow convection and diffusion is subjected to boundary condition constraints and equation closure processing to obtain a set of boundary condition differential equations for liquid flow. It should be noted that since the partial differential equation model of liquid flow convection and diffusion only gives the general form of concentration evolution, its solution is not unique in the absence of boundary and initial constraints, making it impossible to obtain a definite propagation delay. Based on this, the technical solution of this application further sets the global target water and fertilizer concentration scalar as the first type of boundary condition at the beginning, and performs boundary condition constraints and equation closure processing on the partial differential equation model of liquid flow convection and diffusion to obtain a set of boundary condition differential equations for liquid flow. Through the above processing, the propagation process of the concentration shock wave along the path can have a uniquely determined numerical solution.

[0030] More specifically, in a particular example of this application, the partial differential equation model of liquid flow convection and diffusion is used as the main governing equation. Its domain is the longitudinal interval from the inlet to the outlet, covered by a one-dimensional longitudinal spatial vector, as well as the time interval after the injection action. A global target concentration scalar of water and fertilizer is obtained from upstream and applied to the inlet node with a longitudinal coordinate of zero. The concentration at this node is set to be constant at all times after the injection starts, thus constituting the first type of boundary condition at the inlet. This corresponds to the actual working condition where the concentration at this point is directly locked by the mother liquor ratio. Therefore, the constraint is applied in the form of a constant concentration value rather than a concentration gradient. The initial condition is that the concentration at each node in the tank at the start of injection is equal to the background concentration represented by the reference ion concentration values ​​at the inlet and outlet. This background concentration is the residual nutrient level in the tank after the sweet potato seedlings have absorbed nutrients in the previous cycle. The liquid flow at the outlet returns to the storage tank through the return pipe without concentration reflection. Therefore, the first partial derivative of the concentration with respect to the longitudinal coordinate is set to zero, constituting a free outflow boundary. By combining the main governing equation with the three boundary conditions mentioned above, the equations are closed-ended, resulting in the system of differential equations for the flow boundary conditions, as follows: ; in, Vertical axis At the moment The transient ion concentration of the nutrient solution. These are the vertical spatial coordinates extracted from the one-dimensional vertical spatial domain vector. This is the time counted from the start of the injection. The longitudinal hydrodynamic dispersion coefficient, For a one-dimensional advection characteristic velocity scalar, This represents the total length of the hydroponic trough along its length in the geometric dimension vector of the long-distance hydroponic trough. This serves as a scalar for the overall target concentration of water and fertilizer. The background concentration is represented by the baseline ion concentration values ​​at both the beginning and end; the first equation is the main governing equation, and the meaning of each term is consistent with that described in step S2. Its applicable range is limited to the period between the beginning and end and after the start of liquid injection; the second equation... The first type of boundary condition at the head end characterizes the concentration at the head end inlet as locked to the global water and fertilizer target concentration scalar and does not change with the downstream concentration distribution; the third equation As initial conditions, the concentrations at each node from the first to the last at the start of the injection are the residual background concentrations in the tank after absorption by the sweet potato seedlings in the previous cycle; Formula 4 The free outflow boundary condition at the tail end characterizes the fact that the concentration at the liquid outlet at the tail end no longer changes along the longitudinal direction, and the liquid flows directly out of the tank at the current concentration without generating reflection backflow. After the above four equations are combined, the constraints on the unknown concentration function are complete in both time and space dimensions. The system of equations is thus closed, and its solution is uniquely determined within the given domain, forming the basis for subsequent calculations to solve the propagation trajectory of the concentration shock wave.

[0031] In step S32, the concentration shock wave propagation trajectory along the flow path is solved for the flow boundary condition differential equations to obtain the full-time-space concentration response function. It should be noted that since the flow boundary condition differential equations only provide a boundary problem in constraint form and do not give the specific process of concentration increase over time at each longitudinal coordinate, the time of achieving the target at any valve coordinate cannot be directly read. Therefore, the technical solution of this application further solves for the flow boundary condition differential equations by solving for the concentration shock wave propagation trajectory along the flow path to obtain the full-time-space concentration response function. Through the above processing, a continuous concentration distribution covering all longitudinal coordinates and all times in the entire tank can be obtained, providing a data foundation for node-by-node inverse solving of the hysteresis time.

[0032] More specifically, in a concrete example of this application, the Laplace transform is used to solve the differential equations of the liquid flow boundary conditions. After substituting the constraints of the initial concentration constant and the zero concentration gradient at the tail end, and returning to the time domain, an analytical solution for the propagation of the concentration shock wave from the initial to the tail end is obtained. This analytical solution uses the background concentration in the tank before injection as a reference, and superimposes the propagation response of the concentration increment injected at the initial end under the combined action of advection and dispersion, thus obtaining the all-time-space concentration response function, expressed as follows: ; in, The concentration response function across all time and space, representing the vertical axis. At the moment The transient ion concentration of the nutrient solution. This represents the background concentration in the tank before injection. This serves as a scalar for the overall target concentration of water and fertilizer. For a one-dimensional advection characteristic velocity scalar, The longitudinal hydrodynamic dispersion coefficient, For complementary error functions, It is an exponential function with the natural constant as its base; This is half the concentration amplitude that can be increased within the tank. The coordinates are dimensionless. The numerator represents the distance between the observed coordinates and the center of the concentration front, and the denominator represents the width of the front broadened by diffusion at that moment. Therefore, the first complementary error function term represents the main concentration front that moves towards the tail end with the liquid flow and its gradual flattening along the way. The second complementary error function term, which contains an exponential factor, is a correction term introduced by the initial concentration setpoint constraint, which is used to compensate for the reverse transport of diffusion in the upstream direction. The sum of the two terms and the background concentration constitute the complete time history of the concentration at any coordinate rising from the background level and approaching the global water and fertilizer target concentration scalar.

[0033] When the switching of the main circulation pump causes the scalar velocity of the one-dimensional advection to change piecewise and makes it difficult to obtain a closed analytical solution, the finite difference method is used instead. The discrete nodes in the one-dimensional longitudinal spatial domain vector are used as spatial base points. The first-order upwind difference is used for the convection term to suppress numerical oscillations under low flow velocity conditions. The iteration is gradually advanced until the concentration at the tail node tends to stabilize and then terminates. The results are organized into a concentration numerical matrix with the node number as the row and the time step number as the column. This also constitutes the full spatiotemporal concentration response function, where each row is the time history curve of the concentration at the corresponding longitudinal coordinate rising from the background level to near the target concentration.

[0034] In step S33, based on a preset steady-state concentration tolerance threshold coefficient, a node-by-node inverse kinematics of the full-time-space concentration response function is performed to generate a spatiotemporal phase shift feature matrix. It should be noted that since the concentration at each node in the full-time-space concentration response function asymptotically approaches the target value over time without a strict arrival time, a definite hysteresis time cannot be extracted from the continuous time history without setting a judgment criterion. Therefore, the technical solution of this application further performs a node-by-node inverse kinematics of the full-time-space concentration response function based on a preset steady-state concentration tolerance threshold coefficient to generate a spatiotemporal phase shift feature matrix. Through the above processing, the continuous concentration time history can be compressed into a definite hysteresis time corresponding one-to-one with the coordinates of each valve.

[0035] More specifically, in a specific example of this application, a steady-state concentration tolerance threshold coefficient is first set, with a value greater than zero and less than one. This value needs to balance the tolerance range of sweet potato hydroponic seedlings to nutrient concentration and the accuracy of solving the lag time. Under the rapid propagation condition of virus-free sweet potato seedlings, a value of 0.95 is taken, meaning that when the node concentration rises to 95% of the global water and fertilizer target concentration scalar, the nutrient supply of that node is determined to be up to standard. Subsequently, using the coordinates of each discrete node in the one-dimensional vertical spatial domain vector as the traversal object, the concentration time history curve of that node is extracted from the full-time-space concentration response function. The steady-state concentration tolerance threshold coefficient is multiplied by the global water and fertilizer target concentration scalar to obtain the standard concentration threshold. The search is incremented along the time axis from the start of the injection, and the moment when the concentration first does not fall below the threshold is taken as the lag time of that node. For the concentration value matrix obtained by the finite difference method, the time step in which the concentration first crosses the threshold is located in the row corresponding to that node, and the concentration values ​​of that time step and the previous time step are linearly interpolated to improve the resolution of the lag time to below the time step size. The above node-by-node inverse solution process is represented as follows: ; in, Vertical axis The hysteresis time obtained from the inverse solution, The first vector in the one-dimensional vertical spatial domain is the... The vertical coordinates of each node, Let be the concentration value at this node as a function of the all-time-space concentration response, representing the concentration value over time. This is the steady-state concentration tolerance threshold coefficient. This serves as a scalar for the overall target concentration of water and fertilizer. This is the target concentration threshold for this node. It is a subfimum operator used to find the minimum time in the set of times when the concentration is not lower than the threshold, thereby eliminating repeated judgments caused by slight fluctuations in concentration near the threshold during the numerical solution process.

[0036] After completing the inverse kinematics of the hysteresis times of all nodes, the vertical coordinates of each node and their corresponding hysteresis times are paired row by row according to the node number and assembled into a two-dimensional matrix of two columns, resulting in the spatiotemporal phase shift characteristic matrix, as shown below: ; in, The spatiotemporal phase shift characteristic matrix, to The coordinates are the longitudinal coordinates of each node arranged sequentially from the inlet at the beginning to the end. to For the corresponding hysteresis time, This represents the total number of nodes in the one-dimensional longitudinal spatial domain vector. In this matrix, the hysteresis time increases monotonically with the longitudinal coordinate. The first node approaches zero, while the last node has the maximum value for the entire tank. Due to dispersion, the concentration front widens along the path, and the width of the front widening increases according to the square root of time. Therefore, the excess of the time to reach the standard near the first end relative to the advection estimate is relatively large. The hysteresis time increment between adjacent nodes decreases slightly with the coordinate and gradually approaches the ratio of the coordinate difference to the one-dimensional advection characteristic velocity scalar. It does not have a strict linear relationship with the coordinate. This is a quantitative characterization of the sweet potato seedlings at the last end of the long-distance hydroponic tank being in a low-concentration environment, which can be used for subsequent inverse solving based on valve coordinate addressing and injection advance.

[0037] Specifically, in step S4, based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank, the physical space discrete mapping of the hysteresis time along the flow path in the spatiotemporal phase shift feature matrix is ​​performed and decoupled from the distributed control commands to obtain the distributed injection advance time series and the microfluidic pulse volume commands for each node. It should be noted that, since the spatiotemporal phase shift feature matrix only records the hysteresis time required for the concentration to reach the target level at each longitudinal coordinate in data form, its number of rows depends on the spatial discrete density during numerical solution, and does not correspond to the actual number of microfluidic valves deployed along the tank, and the matrix records the hysteresis of the liquid flow reaching each location rather than the time when the valve should act, it cannot be directly executed by the lower-level machine; if it is not mapped to the physical valve coordinates and converted into two types of parameters that can be issued, namely time and volume, then each valve can only act at the same opening degree at the same time, and the concentration mismatch along the flow path still exists. Based on this, the technical solution of this application further uses the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank to perform physical spatial discrete mapping and decoupling of the lag time along the spatiotemporal phase shift feature matrix with distributed control commands to obtain the distributed injection advance time series and the microfluidic pulse volume commands of each node. Through the above processing, the single global concentration decision can be decomposed into timing actions and injection volume commands that match the spatial position of each valve, so that upstream nodes with shorter lag times inject liquid earlier and downstream nodes with longer lag times delay injection liquid.

[0038] Figure 4 Based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic trough, a method for intelligent water and fertilizer management of sweet potato hydroponic seedlings according to an embodiment of this application is used to perform physical spatial discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix and decouple it from the distributed control commands to obtain the distributed injection advance time series and the flowchart of the microfluidic pulse volume commands for each node. For example... Figure 4As shown, step S4 includes: S41, based on the valve coordinates calibrated by equidistant segmentation of the geometric dimension vector of the long-distance hydroponic tank, performing node hysteresis addressing and local volume calculation on the spatiotemporal phase shift feature matrix to obtain the control cluster node hysteresis time set and the local control cluster volume vector; S42, using the maximum hysteresis time of the entire tank segment as the synchronization reference benchmark, performing time delay matrix addressing and inverse solution of the injection advance amount on the control cluster node hysteresis time set to obtain the distributed injection advance amount time series; S43, performing pulse injection volume conversion on the local control cluster volume vector to obtain the microfluidic pulse volume command for each node.

[0039] In step S41, based on the valve coordinates calibrated by equidistant segmentation of the long-distance hydroponic trough geometric dimension vector, node hysteresis addressing and local volume calculation are performed on the spatiotemporal phase shift feature matrix to obtain the control cluster node hysteresis time set and the local control cluster volume vector. It should be noted that, since the spatiotemporal phase shift feature matrix is ​​organized by numerically discrete nodes, its number of rows does not correspond to the actual number of valves deployed along the trough, and the liquid volume controlled by each valve is not yet determined. Based on this, the technical solution of this application further performs node hysteresis addressing and local volume calculation on the spatiotemporal phase shift feature matrix based on the valve coordinates calibrated by equidistant segmentation of the long-distance hydroponic trough geometric dimension vector to obtain the control cluster node hysteresis time set and the local control cluster volume vector. Through the above processing, the corresponding hysteresis time and the amount of fertilizer controlled by each valve can be established.

[0040] More specifically, in a concrete example of this application, the total length of the hydroponic tank, its internal bottom width, and the effective liquid level depth are first extracted from the geometric dimension vector of the long-distance hydroponic tank. The total length of the tank is then linearly divided at equal intervals according to the number of microfluidic valves pre-deployed along the tank. The physical coordinates of each valve on the longitudinal axis are calibrated by successively accumulating the interval step size obtained from the division. The tank is then divided into virtual control clusters equal in number to the total number of microfluidic valves. Each control cluster corresponds to a section of the tank where fertilization is handled by that valve, and the sweet potato seedlings cultivated within that section. The calibration of the valve coordinates is represented as follows: ; in, For the first The longitudinal physical coordinates of a microfluidic valve The valve serial numbers are incremented from the first end to the last end, with values ​​ranging from 1 to... The first valve is located at the inlet at the beginning of the liquid flow path, where the longitudinal coordinate is zero. The valve is located at the beginning. Each valve is located at the upstream end of its respective control cluster. The total length of the tank along the friction line. This represents the total number of microfluidic valves deployed along the channel. The interval step size is the interval obtained by equal-distance division, which is also the slot length of a single virtual control cluster.

[0041] Subsequently, using the longitudinal physical coordinates of each valve as an index, a search is performed in the coordinate column of the spatiotemporal phase shift feature matrix to find a match. The hysteresis time values ​​for the same row are then read and aggregated into a control cluster node hysteresis time set according to valve number. Since the spatial discrete step size of the spatiotemporal phase shift feature matrix is ​​taken as an integer fraction of the valve spacing, each valve coordinate falls on the discrete node coordinates and can be directly addressed. When a valve coordinate is located between two adjacent nodes, the hysteresis time of those two nodes is linearly interpolated based on the coordinate distance and then filled in. Simultaneously, the nutrient solution volume contained in each control cluster is calculated by multiplying the interval step size, internal bottom width, and effective operating liquid level depth. This is arranged by valve number as a local control cluster volume vector, as shown below: ; in, For the first The local volume of a control cluster The width of the bottom inside the tank. The effective liquid level depth is the product of the three factors, which represents the amount of nutrient liquid that needs to be raised to the target concentration within the tank section controlled by the valve. This constitutes the basis for subsequent calculations of the mother liquor injection volume based on mass conservation. In the aforementioned 60m seedling tank, 60 microfluidic valves are deployed at equal intervals along the tank. The interval step length obtained by equal-interval division, i.e., the valve spacing, is 1m, which is exactly twice the spatial discrete step length of 0.5m. Therefore, the coordinates of each valve fall on discrete nodes. The local volume of a single control cluster is 0.01m³, or 10L. The sum of the volumes of all 60 control clusters in the tank is 600L, consistent with the total tank volume. When the cross-section along the tank is consistent, the volumes of each control cluster are equal. However, when the tank has a settling section or a filtration section that causes local cross-section changes, the volume of the corresponding control cluster is calculated separately according to the measured cross-sectional dimensions of that section.

[0042] In step S42, using the maximum hysteresis time of the entire tank section as the synchronization reference, the delay matrix addressing and injection advance time inverse solution are performed on the control cluster node hysteresis time set to obtain the distributed injection advance time series. It should be noted that since the control cluster node hysteresis time set records the lag of the liquid flow reaching the coordinates of each valve, its value increases with the coordinate. If all valves operate at the same time, the concentration attainment time along the flow path will still be inconsistent. Based on this, the technical solution of this application further uses the maximum hysteresis time of the entire tank section as the synchronization reference, and performs delay matrix addressing and injection advance time inverse solution on the control cluster node hysteresis time set to obtain the distributed injection advance time series. Through the above processing, the operating times of each valve can be staggered according to its spatial position, making the concentration attainment time of each control cluster in the entire tank tend to be consistent.

[0043] More specifically, in a concrete example of this application, the hysteresis time values ​​corresponding to the valve numbers in the hysteresis time set of the control cluster nodes are first traversed, and the maximum value is taken as the synchronization reference benchmark for the entire tank section. This maximum value corresponds to the tail valve with the largest longitudinal coordinate, that is, the control cluster where the concentration front arrives latest and the sweet potato seedlings under its jurisdiction receive nutrients last. Using this benchmark as the alignment target, the hysteresis time of each valve is taken out one by one and subtracted from the benchmark. The resulting difference is the time that the valve should act earlier relative to the tail valve. Arranged by valve number, it forms a distributed injection advance time series, as shown below: ; in, For the first The injection advance amount of a microfluidic valve To control the hysteresis time of the cluster node at the valve coordinates, The maximum hysteresis time obtained by traversing all valve numbers is the synchronization reference for the entire tank section. The difference between the two represents the time required for the valve to pre-act in order for its control group and the tail control group to meet the target simultaneously.

[0044] Based on the inverse kinematics results, the injection advance of the tail valve is zero, making it the latest operating node. The lag time of the head valve is close to zero, so its injection advance is close to the maximum lag time of the entire tank section, making it the first to operate. The advance of the valves in the middle section is between these two and decreases with increasing coordinates. Due to the diffusion effect, the concentration front widens along the path, and the lag time increment between adjacent valves decreases slightly with increasing coordinates. Since the injection advance is the difference between the synchronization reference and the lag time, the difference in advance between adjacent valves also decreases slightly with increasing coordinates. The advance distribution of valves becomes denser closer to the tail end. The obtained distributed injection advance time series is stored aligned with the valve number and the local control cluster volume vector for subsequent calculation of pulse volume and generation of trigger times for the superimposed system real-time clock reference.

[0045] In step S43, the local control cluster volume vector is converted into a pulse injection volume to obtain the microfluidic pulse volume command for each node. It should be noted that the local control cluster volume vector only represents the nutrient liquid volume of the tank section under each valve's jurisdiction and has not yet been converted into the mother liquor injection volume that the valve can execute. Therefore, the technical solution of this application further converts the local control cluster volume vector into a pulse injection volume to obtain the microfluidic pulse volume command for each node. Through the above processing, a mother liquor release volume matching the water body under its jurisdiction can be established for each valve.

[0046] More specifically, in a specific example of this application, a mother liquor unit water volume addition coefficient is first established. This coefficient represents the volume of high-concentration mother liquor required to increase the unit volume of nutrient solution from the background concentration to the global water and fertilizer target concentration scalar. Its value is pre-calibrated according to the mass conservation relationship between the original concentration of each ion in the mother liquor tank and the difference between the global water and fertilizer target concentration scalar and the first and last reference ion concentration values, i.e., the concentration difference to be increased is divided by the difference between the original concentration of the mother liquor and the target concentration. Then, the local volume of the control cluster under the jurisdiction of each valve in the local control cluster volume vector is taken out one by one, and multiplied by the addition coefficient. The resulting value is the absolute volume of mother liquor that the valve should release in this injection action. Arranged according to the valve number, it is the microfluidic pulse volume command for each node, as shown below: ; in, For the first Pulse injection volume of a microfluidic valve The mother liquor dosage coefficient per unit volume of water. The local volume of the control cluster governed by the valve is represented by the volume vector of the local control cluster. The product of the two represents the amount of mother liquor required to achieve the target concentration of the nutrient solution within the control cluster. In the above-mentioned seedling tank, the concentration of nitrate nitrogen stock solution in the mother liquor tank is 12000 mg / L, the background concentration before injection is 60 mg / L, and the global target water and fertilizer concentration is 120 mg / L. The mother liquor dosage coefficient per unit water volume is calculated to be approximately 5.05 × 10⁻⁶ by dividing the difference in concentration to be increased by the difference between the stock solution concentration and the target concentration. - Therefore, the pulse injection volume of each valve is approximately 50.5 mL, and the total injection volume of the mother liquor into the 60 valves in the entire tank is approximately 3.03 L. The rated flow rate constant of the microfluidic solenoid valve under the rated differential pressure is 0.25 mL / s. Based on this, the preliminary estimated duration of the single valve action window is approximately 202 s. This duration is much greater than the injection advance difference of approximately 60 s between adjacent valves obtained by the reverse solution in step S42.

[0047] When the cross-section of the tank is consistent and the volume of each control cluster is equal, the pulse injection volume of each valve is also equal, thus ensuring that the sweet potato seedlings cultivated by each control cluster along the tank receive an equal amount of nutrient supplementation. When the tank has a settling section that causes local volume differences, the pulse injection volume of the corresponding control cluster is adjusted proportionally to its actual volume. The obtained microfluidic pulse volume commands for each node and the distributed injection advance time sequence are stored one-to-one according to the valve number. The former determines the mother liquor release volume of each valve, and the latter determines the action time of each valve. Together, they constitute the control parameters required for the lower-level machine to execute phase-controlled injection, which are then called upon when the drive pulse width is calculated and the microfluidic valves evenly distributed along the tank are triggered.

[0048] Specifically, in step S5, microfluidic valves evenly distributed along the tank are triggered based on the distributed injection advance time series. The injection volume of the mother liquor is controlled by microfluidic pulse volume commands at each node, and the pulse width is corrected by feedback from the in-situ monitoring data stream of the hydroponic tank, causing the heterogeneous injected mother liquor to superimpose along the flow path in the low-flow-rate circulating liquid. It should be noted that, since the distributed injection advance time series and the microfluidic pulse volume commands at each node are parameters in the form of relative duration and volume, they are not yet correlated with the absolute clock of the lower-level machine and the opening and closing duration of the solenoid valves, and therefore cannot directly drive the actuator. Furthermore, temperature fluctuations during the seedling stage cause changes in the transpiration intensity of the sweet potato seedlings, resulting in a deviation of the concentration in the tank from the background level used in the calculation. Based on this, the technical solution of this application further uses the distributed injection advance time series to trigger microfluidic valves evenly distributed along the tank, where the injection volume of the mother liquor is controlled by microfluidic pulse volume commands at each node, and the pulse width is corrected by feedback from the in-situ monitoring data stream of the hydroponic tank, causing the heterogeneous injected mother liquor to superimpose along the flow path in the low-flow-rate circulating liquid. Through the above treatment, without increasing the circulation flow rate or changing the microenvironment required by the initial adventitious roots of sweet potato seedlings, each control cluster can simultaneously reach the target concentration, eliminating the mismatch along the path caused by ion toxicity in the first-end seedlings and continuous nutrient deficiency in the last-end seedlings.

[0049] Figure 5 This document describes a method for intelligent water and fertilizer management of sweet potato hydroponic seedlings according to an embodiment of this application. It utilizes a distributed injection advance time-series phase-controlled triggering microfluidic valve evenly distributed along the hydroponic tank. The method involves controlling the mother liquor injection volume with microfluidic pulse volume commands at each node and correcting the pulse width using in-situ monitoring data from the hydroponic tank. This allows the heterogeneous injected mother liquor to superimpose along the flow path in a low-velocity circulating liquid flow. Figure 5 As shown, step S5 includes: S51, based on the system's real-time clock reference, performing absolute trigger time superposition mapping on the distributed injection advance time series to obtain a global phase-controlled trigger timestamp set; S52, based on the reference ion concentration values ​​at the beginning and end ends and the valve's rated flow rate constant, performing error closed-loop fine-tuning and physical drive pulse width calculation on the microfluidic pulse volume commands of each node, and binding the global phase-controlled trigger timestamp set to obtain a multi-node transient physical drive command set; S53, triggering the microfluidic valves evenly distributed along the groove in a phase-dependent manner according to the multi-node transient physical drive command set.

[0050] In step S51, based on the system's real-time clock reference, the distributed injection advance time series is mapped by superimposing absolute trigger times to obtain a global phase-controlled trigger timestamp set. It should be noted that since the distributed injection advance time series only records the advance time of each valve relative to the tail valve and does not contain absolute time information, it cannot be directly invoked by the lower-level machine's timer interrupt. Therefore, the technical solution of this application further maps the distributed injection advance time series by superimposing absolute trigger times based on the system's real-time clock reference to obtain a global phase-controlled trigger timestamp set. Through the above processing, the action times of each valve can fall on the same absolute time axis, thus achieving executable functionality.

[0051] More specifically, in a specific example of this application, the hardware real-time clock of the main control unit is first read to obtain the current absolute time value as the system real-time clock reference. Each valve drive board deployed along the slot is periodically synchronized with the main control unit via the fieldbus, ensuring that the deviation between the local timer of each drive board and this reference is controlled within milliseconds, so as to guarantee that the advance difference between adjacent valves is not overwhelmed by the synchronization error. Then, the injection advance of each valve is extracted one by one from the distributed injection advance time series and added to the system real-time clock reference. The resulting time is the absolute trigger time for the valve to open for injection. These timestamps are collected according to the valve sequence number into a global phase-controlled trigger timestamp set, as shown below: ; ; in, For the first The absolute triggering moment of a microfluidic valve This serves as the reference time for synchronous liquid injection across the entire tank. As the system's real-time clock reference, This represents the injection advance of the valve in the distributed injection advance time series. The maximum injection advance amount obtained by traversing all valve numbers corresponds to the first valve that moves first. The sum of this maximum value and the system real-time clock reference is the reference time for synchronous injection of the entire tank, which is also the opening time of the last valve that moves last. The difference between this reference time and the injection advance amount of a certain valve is the time when the valve should open on the absolute time axis.

[0052] According to the above mapping results, the injection advance of the first valve is the largest and its absolute trigger time is the earliest. The injection advance of the last valve is zero, so its absolute trigger time is equal to the system real-time clock reference and it acts the latest. The trigger times of each valve shift sequentially from the first to the last, and the interval between adjacent trigger times decreases slightly as the coordinate increases. The obtained global phase-controlled trigger timestamp set is written into the timer interrupt register of each driver board according to the valve number, and stored in alignment with the microfluidic pulse volume command of each node according to the valve number. This is used for subsequent calculation of the drive pulse width and binding with the trigger time as a drive command. The microfluidic pulse volume command of each node and the reference ion concentration values ​​of the first and last ends are also sent out along with the control frame.

[0053] In step S52, based on the reference ion concentration values ​​at the beginning and end and the rated flow rate constant of the valve, the microfluidic pulse volume command of each node is fine-tuned with error closed loop and the physical drive pulse width is calculated. A global phased-array trigger timestamp set is then bound to obtain a multi-node transient physical drive command set. It should be noted that since the microfluidic pulse volume command of each node is pre-calculated based on the background concentration at the calculation time according to mass conservation, and there is a time span equivalent to the lead time between the calculation and the actual action of each valve, the increased transpiration of the sweet potato seedlings caused by the rise in greenhouse temperature during this period will further deviate the concentration in the tank. Furthermore, the solenoid valve only accepts drive quantities in the form of on / off duration and cannot directly accept volume quantities. Therefore, the technical solution of this application further performs error closed-loop fine-tuning and physical drive pulse width calculation on the microfluidic pulse volume command of each node based on the reference ion concentration values ​​at the beginning and end and the rated flow rate constant of the valve, and binds a global phased-array trigger timestamp set to obtain a multi-node transient physical drive command set. Through the above processing, the actual release volume of each valve can be dynamically corrected according to the measured deviation of the concentration in the tank, and the two types of parameters, time and duration, can be encapsulated into instructions that can be directly executed by hardware interrupt.

[0054] More specifically, in a specific example of this application, the baseline ion concentration values ​​at the beginning and end of the hydroponic tank are first extracted from the in-situ monitoring data stream after baseline calibration. These values ​​are then subtracted from the background concentration calibration value used when calculating the pulse volume command to obtain the concentration offset error at the current moment. This error is positive when the greenhouse temperature rises and the sweet potato seedlings' transpiration intensifies, and negative when shading and cooling reduce absorption. This error is then fed into an incremental proportional-integral-differential algorithm, where the proportional term responds to the current deviation, the integral term eliminates persistent deviations, and the differential term suppresses rapid changes in deviations. This algorithm calculates the volume fine-tuning compensation term for each valve, and then adds this compensation term element-wise to the microfluidic pulse volume command for each node to obtain the corrected target injection volume. Subsequently, using the valve's rated flow rate constant as the divisor, the corrected target injection volume is converted into the solenoid valve's opening duration, i.e., the physical drive pulse width, as shown below: ; in, For the first The physical drive pulse width of a microfluidic valve This refers to the reference injection volume of the valve in the microfluidic pulse volume command for each node. This is the volume fine-tuning compensation term obtained by using the reference ion concentration values ​​at both ends as feedback quantities and solving it through an incremental proportional-integral-differential algorithm. The numerator represents the flow volume per unit time of the microfluidic solenoid valve under rated differential pressure, as calibrated at the factory, i.e., the valve's rated flow rate constant. The numerator represents the actual volume of mother liquid that the valve should release after error correction, and the denominator linearly converts this volume into the electrical opening duration. Finally, the physical drive pulse width of each valve is bundled with the absolute trigger time of the same valve number in the global phase-controlled trigger timestamp set as key-value pairs, and encapsulated into a multi-node transient physical drive instruction set for the solenoid valve hardware interrupt service routine. Each instruction specifies when a valve opens and for how long it remains open.

[0055] However, the above processing treats the pulse width calculation of each valve as an independent process, and its denominator always uses the factory-calibrated valve rated flow rate constant, which implies the assumption that no matter how many valves are opened at the same time, the transient flow rate at each valve port is always equal to the rated value. In reality, the microfluidic valves deployed along the tank share the same mother liquor supply trunk line. In addition, the flow capacity of the microfluidic solenoid valves is very small. The action window for a single valve to release the amount of mother liquor required for its controlled cluster is as long as hundreds of seconds, while the injection advance difference between adjacent valves is only tens of seconds. The action window duration is several times the trigger interval. The action windows of multiple valves overlap significantly on the time axis and simultaneously draw mother liquor from the shared trunk line. The transient total pumping flow rate in the trunk line is far greater than the level when a single valve works independently, and a transient pressure drop is generated due to friction effect. This pressure drop is transmitted in reverse to the ports of each branch valve, reducing their actual pressure difference, so that the instantaneous actual flow rate of the valve is lower than the rated value. Since the denominator does not perceive or compensate for this flow rate attenuation, the calculated pulse width is too short, and the actual volume of mother liquor injected by each valve is lower than the theoretical design value. Furthermore, because the triggering time of each valve in the middle section is more concentrated than that at the beginning and end, the probability and intensity of synchronous liquid extraction are the highest in the entire tank. The pressure drop of the common pipe is the heaviest, and the deviation of insufficient injection volume is the largest. As a result, a continuous nutrient solution concentration depression zone is formed in the middle of the tank. Sweet potato seedlings in this area are in a state of latent nutrient deficiency for a long time, resulting in slow root development, weak vines, and a lower transplant survival rate than at the beginning and end.

[0056] To eliminate the injection volume deviation caused by treating each valve as an isolated execution unit under multi-valve synchronous operation, the technical solution of this application proposes a second embodiment based on the above processing. First, the degree of overlap of the action windows of each valve on the time axis is quantitatively detected to construct the transient pressure drop coupling matrix between nodes. Then, the rated flow rate of the valve is attenuated by coupling perception according to the matrix and the pulse width is adaptively calculated accordingly, thereby completing the packaging and encapsulation of the drive command.

[0057] Figure 6 This is a flowchart illustrating an intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to an embodiment of this application, which uses adaptive pulse width calculation based on the transient voltage drop coupling matrix between nodes to obtain a multi-node transient physical drive instruction set. (See flowchart for example.) Figure 6 As shown, step S52 includes: S521, based on the absolute trigger time of each node in the global phased trigger timestamp set and the action window interval estimated by the microfluidic pulse volume command of each node, performing node action time domain overlap detection to obtain the transient voltage drop coupling matrix between nodes; S522, based on the reference ion concentration values ​​at the beginning and end and the transient voltage drop coupling matrix between nodes, performing perturbation compensation and coupling sensing adaptive pulse width calculation on the microfluidic pulse volume command of each node to obtain the pulse width vector of each node after coupling compensation; S523, packaging and encapsulating the pulse width vector of each node after coupling compensation and the global phased trigger timestamp set into execution-level instructions oriented towards hardware interrupt service routines to obtain a multi-node transient physical drive instruction set.

[0058] In step S521, based on the absolute trigger time of each node in the global phased-array trigger timestamp set and the action window interval estimated by the microfluidic pulse volume command of each node, the temporal overlap of the nodes is detected to obtain the transient pressure drop coupling matrix between nodes. It should be noted that, given that the microfluidic valves deployed along the tank share the same mother liquor supply trunk line, and the aforementioned pulse width calculation treats each valve as an isolated execution unit and uses the factory-calibrated rated flow rate constant as the denominator, it does not consider the transient pressure drop and flow rate attenuation in the trunk line caused by simultaneous pumping of multiple valves, resulting in the actual injection volume of each valve being lower than the theoretical design value. Therefore, the technical solution of this application further uses the action window interval estimated by the absolute trigger time of each node in the global phased-array trigger timestamp set and the microfluidic pulse volume command of each node to detect the temporal overlap of the nodes and obtain the transient pressure drop coupling matrix between nodes. Through the above processing, the degree of simultaneous pumping of each valve can be converted into a quantifiable coupling coefficient, providing a quantitative basis for subsequent flow rate attenuation compensation.

[0059] More specifically, in a specific example of this application, the absolute trigger time of each node is first extracted from the global phased-array trigger timestamp set as the starting point of the node's action window. Then, the reference injection volume in the microfluidic pulse volume command of each node is divided by the valve's rated flow rate constant to preliminarily estimate the duration of the node's action. The starting point and the duration are added together to obtain the end point of the window, thereby calculating the active action window interval occupied by each node on the time axis. Subsequently, pairwise temporal overlap analysis is performed between all n nodes, that is, the intersection duration of the action windows of any two nodes on the time axis is calculated, and the intersection duration is divided by the duration of the longer window of the two nodes for normalization, thereby obtaining an overlap ratio coefficient with a value between 0 and 1 to eliminate the interference of inconsistent action durations of each node on the determination of the degree of overlap. Finally, the overlap ratio coefficients of all node pairs are assembled into a single value according to the row and column indices. The symmetric matrix, i.e., the transient voltage drop coupling matrix between nodes, is represented as follows: ; in, The coupling coefficient element in the m-th row and n-th column of the transient pressure drop coupling matrix between nodes reflects the intensity of competition between node m and node n for synchronous liquid extraction on the shared mother liquor trunk line. and These are the absolute trigger times of the m-th and n-th nodes in the global phased array trigger timestamp set, respectively. and These are the durations of the action windows for the m-th and n-th nodes, respectively, estimated based on the microfluidic pulse volume commands and the valve's rated flow rate constant at each node. and These represent the active action window intervals of the two nodes on the timeline. The intersection length operator for two time intervals. The longer of the two action windows is taken as the normalization benchmark. The numerator represents the actual duration for which the two nodes simultaneously begin pumping, and the denominator converts this duration into a dimensionless ratio to make nodes with different injection volumes comparable. It should be noted that the diagonal elements of this matrix are always 1, meaning the node completely overlaps with itself. The closer the off-diagonal elements are to 1, the closer the corresponding two nodes are to a completely synchronized working state, and the more severe the superimposed pressure drop on the common pipe. By constructing this matrix, the synchronous pumping competition relationship between valves is characterized as a quantifiable coupling coefficient, providing a quantitative basis for subsequent flow rate attenuation compensation.

[0060] For example, in a seedling tray used for rapid propagation of virus-free sweet potato seedlings, 60 microfluidic valves are equidistantly deployed along the tray. The injection advance, obtained through reverse engineering, decreases from the first end to the last end. The difference in advance between adjacent valves is approximately 60 seconds. The duration of each valve's operation, estimated based on the volume of its control cluster and the mother liquor addition coefficient, is approximately 202 seconds, more than three times the former. Therefore, during the opening of any valve, its three adjacent valves upstream and downstream are all open, resulting in a large overlap of the operation windows on the time axis. According to the above formula, the overlap ratio coefficients with the first, second, and third adjacent valves are approximately 0.70, 0.41, and 0.11, respectively. There is no overlap with valves more than four valves apart, and the coefficient is 0. Mid-section valves have the aforementioned overlapping neighborhoods on both their upstream and downstream sides, while the first and last end valves only have adjacent valves on one side. Therefore, the number of non-zero, non-diagonal elements in their rows is approximately half that of the mid-section valves. The resulting transient pressure drop coupling matrix between nodes thus presents a high-value clustering zone centered on the middle node. This clustering zone spatially corresponds to the concentration depression zone formed in the middle of the tank and the area where the sweet potato seedling root development is slow. Based on this, the phenomenon of weak growth in the middle, which could only be discovered after the fact by observing the seedling condition, can be located and quantified in the form of matrix elements before the command is issued, for use in subsequent calculation of the comprehensive coupling load factor and correction of the effective flow rate of the valve.

[0061] In step S522, based on the reference ion concentration values ​​at the beginning and end points and the coupling matrix of transient pressure drop between nodes, the microfluidic pulse volume command of each node is perturbed and adaptively pulsed with coupling sensing to obtain the pulse width vector of each node after coupling compensation. It should be noted that, given that the aforementioned pulse width calculation always uses the factory-calibrated valve rated flow rate constant as the denominator, it does not compensate for the pressure drop in the main pipe and the attenuation of the port flow rate caused by simultaneous pumping of multiple valves, resulting in the actual injection volume of each valve being lower than the theoretical design value, and the middle section valves having the most severe deviation due to the concentrated triggering time. Based on this, the technical solution of this application further performs perturbation compensation and adaptive pulse width calculation of the microfluidic pulse volume command of each node based on the reference ion concentration values ​​at the beginning and end points and the coupling matrix of transient pressure drop between nodes to obtain the pulse width vector of each node after coupling compensation. Through the above processing, the opening time of each valve can be adaptively extended according to the intensity of synchronous pumping competition, so that the actual released mother liquid volume returns to the theoretical design value.

[0062] More specifically, in a specific example of this application, firstly, the reference ion concentration values ​​at the beginning and end are extracted as environmental disturbance feedback signals as described in step S52. A volume fine-tuning compensation term is generated using an incremental proportional-integral-differential algorithm and superimposed element-by-element with the microfluidic pulse volume commands of each node to obtain the corrected target injection volume. Subsequently, instead of using a constant valve rated flow rate constant as the divisor, the transient pressure drop coupling matrix between nodes is first summed row-wise to obtain the comprehensive coupling load factor faced by each node. This factor physically characterizes the total pumping competition intensity experienced by the entire shared trunk line during the node's action window. Using this load factor, the valve rated flow rate is mapped exponentially to calculate the effective transient flow rate of each node under real multi-valve cooperative operating conditions. Finally, the corrected target volume is divided by this effective transient flow rate to obtain the pulse width vector of each node after coupling compensation, as shown below: ; in, This represents the actual driving pulse width of the m-th node in the pulse width vector of each node after coupling compensation. The reference injection volume parameters for the m-th node in the microfluidic pulse volume command for each node are defined. The volume fine-tuning compensation term is calculated using an incremental proportional-integral-differential algorithm, with the reference ion concentrations at both ends as feedback values. The reference ion concentration values ​​are at both the beginning and end. This represents the rated flow rate constant of the microfluidic solenoid valve under ideal, uncoupled operating conditions. This represents the coupling coefficient element in the m-th row and n-th column of the transient voltage drop coupling matrix between nodes. This represents the total number of microfluidic valves deployed along the channel. The comprehensive coupling load factor is extracted and accumulated from the m-th row of the transient pressure drop coupling matrix between nodes. The larger the value, the more intense the competition for simultaneous liquid pumping faced by that node. The pressure drop and velocity decay index, determined by the characteristics of the mother liquor trunk pipeline, is obtained through pre-calibration based on pipe diameter, pipe length, and fluid viscosity. The value falls within the range of greater than zero and less than one. In the aforementioned mother liquor trunk pipeline with an inner diameter of 8 mm and a total length of 65 m, the calibrated value is 0.28, characterizing the nonlinear power-law relationship between hydraulic pressure loss and flow rate. The numerator represents the volume of mother liquor that should be released at this node after environmental disturbance compensation, and the denominator... This represents the effective transient flow rate of the node under multi-valve coordinated operation, which decreases according to a power law as the overall coupling load factor increases. It should be noted that when the overall coupling load factor is greater than 1, that is, other nodes are operating synchronously with the current node, the effective flow rate in the denominator is attenuated to below the rated value, and the pulse width is adaptively lengthened accordingly, accurately compensating for the injection volume lost due to the common pipe pressure drop.

[0063] In a seedling trough used for rapid propagation of virus-free sweet potato seedlings, the overlap ratios of the valves in the middle section and the three adjacent valves on each side are approximately 0.70, 0.41, and 0.11, respectively. The overall coupling load factor obtained by summing these values ​​row by row is approximately 3.44. Substituting this into the above formula, the effective transient flow velocity decreases to about 70% of the rated value, and the corresponding pulse width increases by about 40% compared to the uncompensated state. The valves at the beginning and end, because they only have overlapping neighborhoods on one side, have an overall coupling load factor of approximately 2.22, with the effective transient flow velocity decreasing to 80% of the rated value and the pulse width increasing by about 25%, a significantly smaller increase than that of the valves in the middle section. Under the condition that a single valve operates independently without any synchronous competition, the overall coupling load factor equals one, and the above formula degenerates into the same calculation result as the uncompensated state, thus ensuring the backward compatibility of the improved mechanism. As a result, the volume of mother liquor released by each valve in the same phase-controlled injection tends to be consistent, transforming the latent defect of insufficient systematic injection of each valve in the central phase-dense zone of the long-distance hydroponic tank due to the pressure drop of the common pipe into a quantifiable and compensable explicit engineering parameter. This eliminates the root cause of long-term nutrient deficiency in sweet potato seedlings in this area from a control perspective, and the persistent concentration depression in the central tank section disappears. The rate of adventitious root formation of sweet potato seedlings in this area tends to be consistent with that at both ends.

[0064] In step S523, the pulse width vectors of each node after coupling compensation and the global phase-controlled trigger timestamp set are packaged and encapsulated into execution-level instructions for hardware interrupt service routines to obtain a multi-node transient physical drive instruction set. It should be noted that, given that the pulse width vectors of each node after coupling compensation and the global phase-controlled trigger timestamp set belong to two independent sets of data, and a corresponding relationship has not yet been established between them by node, the solenoid valve hardware interrupt service routine cannot determine when to open the valve and for how long. Based on this, the technical solution of this application further packages and encapsulates the pulse width vectors of each node after coupling compensation and the global phase-controlled trigger timestamp set into execution-level instructions for hardware interrupt service routines to obtain a multi-node transient physical drive instruction set. Through the above processing, the pulse width parameters with embedded common-channel voltage drop feedforward compensation and the absolute trigger time can be simultaneously sent to the actuator.

[0065] More specifically, in a specific example of this application, a fixed-length instruction frame structure is written node by node, using the node number as the key and a tuple consisting of the absolute trigger time and the pulse width as the value. A verification field is attached to each frame for the driver board to check data integrity. Subsequently, each instruction frame is sent to the corresponding valve driver board deployed along the channel via the fieldbus. The driver board writes the absolute trigger time into the timing comparison register and the pulse width into the timing reload register. When the local timer matches the synchronized system clock reference, a hardware interrupt is triggered. The interrupt service routine sets the solenoid valve drive pin and resets it when the pulse width is exhausted, thus maintaining the determinism of the trigger time regardless of fluctuations in the host computer's computational load. After this encapsulation, the output driver instruction set embeds feedforward compensation information for the common pipe pressure drop effect caused by multi-valve synchronous competition in the pulse width of each node. This ensures that the volume of mother liquid released by each valve at the physical level under actual collaborative working conditions approaches the theoretical design amount, thereby preventing the uniformity effect of the downstream phase-controlled injection process from being affected by the degradation caused by pipeline hydraulic coupling interference.

[0066] In a seedling trough for rapid propagation of virus-free sweet potato seedlings, the absolute trigger time corresponding to the first valve in the encapsulated instruction set is the earliest. Its pulse width is extended by approximately 25% according to the comprehensive coupling load factor of approximately 2.22 due to the overlapping neighborhood on only one side. The absolute trigger times corresponding to the valves in the middle section are more concentrated, and their pulse widths have been extended according to the comprehensive coupling load factor. The absolute trigger time corresponding to the valve at the tail end is the latest. Each drive board independently completes valve opening and closing according to the parameters within the frame. After heterogeneous injection, the mother liquor is superimposed along the flow path in the low-flow-rate circulating liquid flow. The middle section control cluster no longer experiences insufficient mother liquor injection due to the short valve opening time. The concentration of each control cluster in the entire trough reaches the target level at the same time. The rate of adventitious root development and the stem thickness of the sweet potato seedlings along the trough tend to be consistent, and the uniformity of the transplant survival rate of the same rapid propagation batch is maintained.

[0067] In step S53, the microfluidic valves evenly distributed along the tank are triggered out of phase according to the multi-node transient physical drive instruction set. It should be noted that, since the multi-node transient physical drive instruction set is still a set of control parameters stored in a register, it has not yet been converted into the actual distribution of the mother liquor within the tank. Based on this, the technical solution of this application further triggers the microfluidic valves evenly distributed along the tank out of phase according to the multi-node transient physical drive instruction set. Through the above processing, the mother liquor released by each valve at different times can be superimposed along the flow path in the low-flow-rate circulating liquid, thus ensuring that the entire tank simultaneously meets the standards.

[0068] More specifically, in a specific example of this application, each valve drive board continuously compares the local timer with the absolute trigger time within the instruction frame. When the comparison matches, a hardware interrupt is triggered and the solenoid valve drive pin is set. The valve draws high-concentration mother liquor from the common mother liquor trunk and injects it into the tank segment where its control cluster is located through the bypass tube. When the local timer accumulates to the pulse width carried in the frame, the drive pin is reset and the valve is closed. Since the absolute trigger time of the first valve is the earliest and the last valve is the latest, each valve opens in a phase-dependent order from the first to the last. At most, some valves are open at the same time. The mother liquor is injected in a pulsed manner without changing the operating level of the main circulation pump. The one-dimensional advection characteristic velocity scalar is maintained within the shear tolerance range of the primary adventitious roots of the sweet potato seedlings.

[0069] The mother liquor pulses injected by each valve flow laterally towards the tail end with the circulating liquid and widen along the path under the effect of dispersion. The concentration peak of the upstream pulse injected earlier has already leveled off when it reaches the downstream tank section. At this moment, the valves of the tank section are opened and local mother liquor is added. The concentration increases of the two are superimposed within the tank section. Since the opening time of each valve is obtained by inversely solving its lag time relative to the maximum lag time of the entire tank section, the concentration rise process after superposition at each point is aligned in time. The concentration of each control cluster in the entire tank reaches the level set by the global water and fertilizer target concentration scalar at the same moment, forming a nutrient solution distribution state with the concentration variance along the longitudinal axis tending to be minimal, which can be represented as follows: ; in, The longitudinal coordinate after phase-controlled liquid injection is completed. The actual nutrient solution concentration at that location, This represents the background concentration in the tank before injection. This represents the total number of microfluidic valves deployed along the channel. For valve serial numbers, For the first The longitudinal physical coordinates of each valve. This represents the pulse width of the valve in the pulse width vector of each node after coupling compensation. The concentration increment response function of the mother liquor pulse after advection and dispersion mixing characterizes the amount of mother liquor released by the valve at its coordinate position relative to the coordinate position by the pulse width. The concentration increment contributed by the location The summation term represents the total target concentration of water and fertilizer as a scalar value, and represents the mother liquor released by all valves in the coordinate system. The sum of the incremental concentrations at each point, along with the background concentration, approaches the target concentration, indicating that the concentration mismatch along the route has been eliminated.

[0070] Furthermore, an intelligent water and fertilizer management system for sweet potato hydroponic seedlings is also provided.

[0071] Figure 7This is a block diagram of an intelligent water and fertilizer management system for hydroponic sweet potato seedlings according to an embodiment of this application. Figure 7 As shown, an intelligent water and fertilizer management system 100 for sweet potato hydroponic seedlings according to an embodiment of this application includes: a nutrient deficiency analysis module 110, used to perform nutrient deficiency analysis on the collected in-situ monitoring data stream of the hydroponic trough based on the nutrient demand model of sweet potato hydroponic seedlings to obtain a global target concentration scalar of water and fertilizer; a liquid flow transport modeling module 120, used to perform one-dimensional spatial discretization and advection dynamic mapping on the geometric dimension vector of the long-distance hydroponic trough based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic trough to construct a liquid flow convection diffusion partial differential equation model; and a spatiotemporal phase shift inversion module 130, used to inject the global target concentration scalar of water and fertilizer as the initial excitation into the liquid flow convection diffusion partial differential equation model to analyze the nutrient solution concentration. The propagation delay of the shock wave along the path is numerically inverted to obtain the spatiotemporal phase shift feature matrix; the distributed command decoupling module 140 is used to perform physical space discrete mapping and distributed control command decoupling on the path lag time in the spatiotemporal phase shift feature matrix based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank to obtain the distributed injection advance time series and the microfluidic pulse volume command of each node; the phase-controlled injection execution module 150 is used to phase-control trigger the microfluidic valves uniformly distributed along the tank based on the distributed injection advance time series, wherein the injection volume of mother liquor is controlled by the microfluidic pulse volume command of each node and the pulse width is corrected by the feedback of the in-situ monitoring data stream of the hydroponic tank so that the heterogeneous injected mother liquor is superimposed along the path in the low-flow-rate circulating liquid flow.

[0072] As described above, the intelligent water and fertilizer management system 100 for sweet potato hydroponic seedlings according to embodiments of this application can be implemented in various types of computing devices or control units, such as servers deployed in the seedling farm control center, or edge controllers installed at the hydroponic trough site. In one possible implementation, the intelligent water and fertilizer management system 100 for sweet potato hydroponic seedlings according to embodiments of this application can be integrated into the computing device as a software module and / or hardware module. For example, the intelligent water and fertilizer management system 100 for sweet potato hydroponic seedlings can be a resident water and fertilizer phase-controlled injection scheduling service in the operating system of the computing device. This software module is configured to perform baseline calibration and nutrient deficiency analysis of the in-situ monitoring data stream of the hydroponic trough, construction of the partial differential equation model of liquid flow convection diffusion and inversion of concentration shock wave propagation time delay, decoupling of valve coordinate addressing and injection advance and pulse volume commands, error closed-loop solution of driving pulse width and trigger timestamp binding, and phase-controlled out-of-phase triggering of microfluidic valves and control of mother liquor injection volume, or it can be a dedicated phase-controlled injection management program developed for the computing device. Of course, the intelligent water and fertilizer management system 100 for sweet potato hydroponic seedlings can also be one of the many hardware modules of the computing device or control unit, or it can be embedded in the field programmable gate array circuit to accelerate the numerical inversion of concentration propagation delay and the comparison and determination of multi-node triggering time with hardware parallel logic, or it can be a dedicated integrated circuit for calculating the injection advance amount and generating valve drive pulse width for a specific application.

[0073] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for intelligent water and fertilizer management of hydroponic sweet potato seedlings, characterized in that, include: S1. Based on the nutrient requirement model of sweet potato hydroponic seedlings, nutrient deficit analysis is performed on the collected in-situ monitoring data stream of the hydroponic tank to obtain the global target water and fertilizer concentration scalar. The in-situ monitoring data stream of the hydroponic tank includes the growth stage category label of sweet potato hydroponic seedlings, ion concentration signals at the beginning and end of the nutrient solution, conductivity signal, liquid temperature signal, and main circulation pump flow rate parameters, including: S11. Based on the agronomic nutrition knowledge graph of sweet potato hydroponic seedlings, growth stage state mapping and ion concentration baseline calibration are performed on the in-situ monitoring data stream of the hydroponic tank to obtain the standard nutrient requirement concentration vector and the baseline ion concentration values ​​at the beginning and end; S12. Dynamic nutrient deficit calculation is performed on the standard nutrient requirement concentration vector and the baseline ion concentration values ​​at the beginning and end to obtain the system nutrient deficit characteristics; S13. State value optimization and action decision-making are performed on the system nutrient deficit characteristics to obtain the global target water and fertilizer concentration scalar. S2. Based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic tank, the geometric dimension vector of the long-distance hydroponic tank is discretized in one dimension and mapped by advection dynamics to construct a partial differential equation model of liquid flow convection diffusion. S3. Using the global target concentration of water and fertilizer as the scalar value of the first-end excitation injection liquid flow convection diffusion partial differential equation model, numerical inversion is performed on the propagation delay of the nutrient solution concentration shock wave along the path to obtain the spatiotemporal phase shift characteristic matrix. S4. Based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank, the physical space discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix and the decoupling of the distributed control command are performed to obtain the distributed injection advance time series and the microfluidic pulse volume command of each node. S5. A microfluidic valve with phase-controlled triggering based on distributed injection advance time series is uniformly distributed along the tank. The injection volume of the mother liquor is controlled by the microfluidic pulse volume command of each node, and the pulse width is corrected by the feedback of the in-situ monitoring data stream of the hydroponic tank so that the heterogeneous injected mother liquor is superimposed along the flow in the low-flow-rate circulating liquid flow.

2. The intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to claim 1, characterized in that, Step S2 includes: S21. The cross-sectional area is calculated and the linear velocity is converted from the geometric dimension vector of the long-distance hydroponic tank and the flow velocity parameters of the main circulation pump to obtain the one-dimensional advection characteristic velocity scalar and the effective flow cross-sectional area of ​​the hydroponic tank. S22. Transform the geometric dimension vector of the long-distance hydroponic trough with the effective flow cross-sectional area of ​​the hydroponic trough into a one-dimensional longitudinal spatial domain vector by performing a flow dynamics parameter transformation. S23. Assemble the convection-diffusion operator by combining the one-dimensional longitudinal spatial domain vector and the one-dimensional advection characteristic velocity scalar to construct a partial differential equation model of liquid flow convection-diffusion.

3. The intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to claim 1, characterized in that, Step S3 includes: S31. Set the global water and fertilizer target concentration scalar as the first type of boundary condition at the beginning, and perform boundary condition constraints and equation closure processing on the liquid flow convection diffusion partial differential equation model to obtain the liquid flow boundary condition differential equation set. S32. Solve the differential equations for the liquid flow boundary conditions to obtain the concentration shock wave propagation trajectory along the path to obtain the all-time-space concentration response function. S33. Based on the preset steady-state concentration tolerance threshold coefficient, perform a node-by-node underfimum inverse solution of the full-time concentration response function to generate a spatiotemporal phase shift feature matrix.

4. The intelligent water and fertilizer management method for hydroponic sweet potato seedlings according to claim 1, characterized in that, Step S4 includes: S41. Based on the valve coordinates calibrated by equidistant segmentation of the geometric dimension vector of the long-distance hydroponic tank, node hysteresis addressing and local volume calculation are performed on the spatiotemporal phase shift feature matrix to obtain the control cluster node hysteresis time set and local control cluster volume vector. S42. Using the maximum hysteresis time of the entire tank section as the synchronization reference benchmark, perform time delay matrix addressing and injection advance inverse solution on the hysteresis time set of control cluster nodes to obtain the distributed injection advance time series. S43. Perform pulse injection volume conversion on the local control cluster volume vector to obtain the microfluidic pulse volume command for each node.

5. The intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to claim 1, characterized in that, Step S5 includes: S51. Based on the system's real-time clock reference, the absolute trigger time superposition mapping of the distributed injection advance time series is performed to obtain a global phase control trigger timestamp set. S52. Based on the reference ion concentration values ​​at the beginning and end and the rated flow rate constant of the valve, the error closed-loop fine adjustment and physical drive pulse width calculation of the microfluidic pulse volume command of each node are performed, and the global phase control trigger timestamp set is bound to obtain the multi-node transient physical drive command set. S53. Microfluidic valves uniformly distributed along the slots are triggered by out-of-phase triggering according to the multi-node transient physical drive instruction set.

6. The intelligent water and fertilizer management method for sweet potato hydroponic seedlings according to claim 5, characterized in that, Step S52 includes: S521. Based on the absolute trigger time of each node in the global phased trigger timestamp set and the action window interval estimated by the microfluidic pulse volume command of each node, the temporal overlap of each node is detected to obtain the transient voltage drop coupling matrix between nodes. S522. Based on the reference ion concentration values ​​at the beginning and end and the coupling matrix of transient voltage drop between nodes, the microfluidic pulse volume command of each node is perturbed and coupled to adaptive pulse width calculation to obtain the pulse width vector of each node after coupling compensation. S523. After coupling compensation, the pulse width vector of each node and the global phase control trigger timestamp set are packaged and encapsulated into execution-level instructions for hardware interrupt service routines to obtain a multi-node transient physical drive instruction set.

7. An intelligent water and fertilizer management system for hydroponic sweet potato seedlings, characterized in that, include: The nutrient deficit analysis module is used to analyze the nutrient deficit of the collected in-situ monitoring data stream of the hydroponic trough based on the nutrient requirement model of sweet potato hydroponic seedlings to obtain the global target water and fertilizer concentration scalar. The in-situ monitoring data stream of the hydroponic trough includes the growth stage category label of sweet potato hydroponic seedlings, ion concentration signals at the beginning and end of the nutrient solution, conductivity signal, liquid temperature signal, and main circulation pump flow rate parameters. The module includes: S11, mapping the growth stage state and calibrating the ion concentration baseline of the in-situ monitoring data stream of the hydroponic trough based on the agronomic nutrition knowledge graph of sweet potato hydroponic seedlings to obtain the standard nutrient requirement concentration vector and the baseline ion concentration values ​​at the beginning and end; S12, performing dynamic nutrient deficit calculation on the standard nutrient requirement concentration vector and the baseline ion concentration values ​​at the beginning and end to obtain the system nutrient deficit characteristics; S13, performing state value optimization and action decision-making on the system nutrient deficit characteristics to obtain the global target water and fertilizer concentration scalar. The fluid transport modeling module is used to perform one-dimensional spatial discretization and advection dynamic mapping on the geometric dimension vector of a long-distance hydroponic tank based on the main circulation pump velocity parameters in the in-situ monitoring data stream of the hydroponic tank, so as to construct a partial differential equation model of fluid convection and diffusion. The spatiotemporal phase shift inversion module is used to take the global water and fertilizer target concentration scalar as the first-end excitation injection liquid flow convection diffusion partial differential equation model, and to perform numerical inversion on the propagation delay of the nutrient solution concentration shock wave to obtain the spatiotemporal phase shift characteristic matrix. The distributed command decoupling module is used to perform physical space discrete mapping of the hysteresis time along the spatiotemporal phase shift feature matrix and decouple it from the distributed control command based on the valve coordinates calibrated by the geometric dimension vector of the long-distance hydroponic tank, so as to obtain the distributed injection advance time series and the microfluidic pulse volume command of each node. The phase-controlled liquid injection execution module is used to trigger microfluidic valves evenly distributed along the tank based on the phase-controlled time series of distributed liquid injection advance. The injection volume of mother liquor is controlled by the microfluidic pulse volume command of each node, and the pulse width is corrected by the feedback of the in-situ monitoring data stream of the hydroponic tank so that the heterogeneous injected mother liquor is superimposed along the flow in the low-flow-rate circulating liquid flow.

8. The intelligent water and fertilizer management system for hydroponic sweet potato seedlings according to claim 7, characterized in that, The phase-controlled liquid injection execution module includes: The trigger time mapping unit is used to perform absolute trigger time superposition mapping on the distributed injection advance time series based on the system real-time clock reference to obtain a global phase control trigger timestamp set; The driving pulse width calculation unit is used to perform error closed-loop fine-tuning and physical driving pulse width calculation on the microfluidic pulse volume commands of each node based on the reference ion concentration values ​​at the beginning and end and the valve rated flow rate constant, and binds the global phase control trigger timestamp set to obtain the multi-node transient physical driving command set. The valve phase triggering unit is used to trigger microfluidic valves evenly distributed along the groove in phase according to the multi-node transient physical drive instruction set.

Images