Landscape plant precise maintenance control method based on multi-source perception and growth model

Abstract

The invention discloses a landscape plant precise maintenance control method based on multi-source perception and a growth model, particularly relates to the technical field of automatic control systems, and is used for solving the problem that a set value in an existing landscape plant maintenance control system is kept constant and cannot adapt to dynamic growth requirements of plants. The optimal environment set value of the current growth stage is calculated by collecting growth environment parameters and plant physiological state parameters, a multivariable time sequence causal network is constructed based on the parameters, the time sequence evolution entropy is calculated, and the change trend of the set value is monitored. Identifying a conflict relationship among different maintenance targets, analyzing a coordination priority through a key causal path to generate a coordinated maintenance strategy, and generating a set value smooth switching track when a time sequence evolution entropy exceeds an adaptive threshold value and a set value change trend is consistent with a phenological period prediction direction; and finally, the controller is adjusted to output an instruction to drive an execution mechanism to carry out maintenance operation according to the updated set value, and optimization of the control system on the whole life cycle of plant growth is achieved.

Description

Technical Field

[0001] This invention relates to the field of automated control system technology, and more specifically, to a method for precise maintenance and control of landscape plants based on multi-source sensing and growth models. Background Technology

[0002] In automated maintenance systems for landscape plants, existing technologies collect environmental data by deploying multiple sensors and use plant growth models to calculate optimal growth environment parameters as the target setpoints for the control system. A closed-loop control architecture is employed, with the controller driving the actuators to track and maintain the system state at these setpoints. The core of these existing technologies lies in optimizing controller performance to achieve precise tracking and stable maintenance of the setpoints.

[0003] The shortcomings of existing technologies are that their control architecture is based on the premise that the set value is kept constant for optimization. However, the optimal set value output by the plant growth model at different growth stages has time-varying characteristics. This contradiction between the static preset mode of the control target and the dynamic optimization needs of plant growth makes it impossible for the control system to adaptively follow the dynamic growth needs of the plant throughout its entire life cycle, thus limiting the accuracy and adaptability of maintenance control. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for precise maintenance and control of landscape plants based on multi-source sensing and growth model to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A precise maintenance and control method for landscape plants based on multi-source sensing and growth models includes: S1. Continuously collect environmental parameters of landscape plants and physiological state parameters of plants through multi-source sensors; S2. Input the growth environment parameters and plant physiological state parameters into the plant growth model, calculate the optimal environmental settings for the current growth stage, and generate phenological prediction results simultaneously. S3. Based on growth environment parameters and plant physiological state parameters, a causal network of multivariate time series is constructed using a causal discovery algorithm. The temporal evolution entropy of the causal network topology is calculated, and the changing trend of the optimal environmental setpoint is monitored. S4. Identify conflicting relationships between different maintenance objectives based on the causal network topology, and coordinate the priority of objectives by analyzing key causal paths to generate a coordinated maintenance strategy. S5. When the time sequence evolution entropy exceeds the adaptive threshold and the trend of the optimal environmental setpoint is consistent with the direction of change indicated by the phenological period prediction results, a smooth switching trajectory of the setpoint is generated based on the phenological period prediction results and the coordinated maintenance strategy. S6. Based on the set value, smoothly switch the trajectory, adjust the controller's output command, and drive the actuator to perform maintenance work according to the updated set value.

[0006] Furthermore, in S1, the multi-source sensors include a temperature sensor, a humidity sensor, a light sensor, a leaf surface temperature sensor, and a stem diameter sensor; The growth environment parameters were collected by temperature, humidity and light sensors; Plant physiological parameters were collected by leaf surface temperature sensors and stem diameter sensors; During the data collection process, the multi-source sensors use a unified time reference to collect data synchronously, ensuring that the growth environment parameters and plant physiological state parameters remain consistent over time.

[0007] Furthermore, in S2, the plant growth model calculates the optimal environmental setpoints based on growth environment parameters and plant physiological state parameters through a preset growth stage mapping relationship; And generate phenological prediction results using a phenological prediction algorithm; The growth stage mapping relationship is pre-established based on the plant growth characteristics, and the phenological period prediction algorithm is trained based on historical growth data.

[0008] Furthermore, the phenological period prediction algorithm establishes a prediction model by analyzing historical growth environment parameters and plant physiological state parameters with phenological period observation records; inputs current growth environment parameters and plant physiological state parameters into the prediction model, calculates phenological period state values ​​through multiple regression equations; and maps phenological period state values ​​to preset phenological period stages to generate phenological period prediction results.

[0009] Furthermore, in S3, the causal discovery algorithm processes time-series data of growth environment parameters and plant physiological state parameters, establishes causal relationships between parameters, and constructs a causal network. Calculate the temporal evolution entropy based on the changes in node connectivity in a causal network; Meanwhile, the trend of change is monitored by analyzing the numerical changes of the optimal environmental setpoint over time. The calculation of temporal evolution entropy is based on the time series change rate of the node centrality index in the causal network topology. The change trend of the optimal environmental setting value is obtained by numerical difference calculation at its continuous time points.

[0010] Furthermore, the node centrality index is node degree centrality; the node degree centrality time series is obtained by using a sliding time window, and the rate of change of node degree centrality between adjacent time windows is calculated; the standard deviation of all node degree centrality change rate series is calculated to obtain the temporal evolution entropy.

[0011] Furthermore, in S4, conflict relationships are identified based on the connection strength and direction between nodes in the causal network topology; Extract the key causal paths connecting nodes with different maintenance objectives in the causal network topology; For conflicting relationships, the intensity of causal influence on each key causal path is analyzed, and the maintenance objectives with conflicting relationships are prioritized according to the intensity of causal influence on the key causal paths, thereby generating a coordinated maintenance strategy.

[0012] Furthermore, the extraction of key causal paths includes: identifying nodes representing different maintenance goals, searching for all paths representing nodes representing different maintenance goals in the causal network topology; calculating the causal influence strength of the path based on the product of the connection weights between adjacent nodes on the path; and selecting the path with the greatest causal influence strength as the key causal path.

[0013] Furthermore, in S5, the degree of system state change is judged by comparing the temporal evolution entropy with the adaptive threshold dynamically adjusted based on historical data. At the same time, the rationality of the growth stage transition is verified by analyzing the consistency between the changing trend of the optimal environmental setpoint and the changing direction indicated by the phenological period prediction results. When both the degree of change in system state and the rationality of the transition of growth stage are met, the target growth stage is determined based on the phenological period prediction results. Combined with the target priority in the coordinated maintenance strategy, a smooth transition function is used to generate a smooth switching trajectory of the set value that changes continuously in the time dimension.

[0014] Furthermore, in S6, the setpoint smooth switching trajectory is discretized into a sequence of target setpoints for multiple control cycles; Based on the deviation between the target setpoint sequence and the current actual measured value, the controller's output command is calculated using a control algorithm. The controller's output commands drive the actuator to perform actions, causing environmental parameters to track the target setpoint sequence; The actuators include irrigation equipment, shading equipment, and ventilation equipment, which respectively regulate environmental parameters such as moisture, light, and temperature.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By deeply integrating multi-source sensing with growth models, dynamic adaptive optimization of the landscape plant maintenance control system is achieved. Based on the growth environment parameters and plant physiological state parameters continuously collected by multi-source sensors, the optimal environmental setpoint for the current growth stage is calculated in real time through the plant growth model, and phenological prediction results are generated simultaneously. This allows the control target to dynamically adjust in accordance with the natural rhythm of plant growth. By constructing a causal network of multivariate time series and calculating the temporal evolution entropy, key nodes of growth stage transitions can be keenly captured. At the same time, the consistency between the changing trend of the optimal environmental setpoint and the direction of phenological prediction is monitored, ensuring the reliability of the growth stage transition judgment. This breaks through the limitation of keeping the setpoint constant in traditional control systems, enabling the control system to adaptively match the dynamic needs of the entire plant life cycle, significantly improving the accuracy and environmental adaptability of maintenance control.

[0016] 2. By identifying conflicting relationships between maintenance objectives through causal network topology, and achieving intelligent coordination of objective priorities based on key causal path analysis, maintenance strategies that fully consider plant physiological characteristics are generated. When the time sequence evolution entropy exceeds the adaptive threshold and the trend of setpoint change is consistent with the phenological period prediction, the system generates a smooth setpoint switching trajectory based on the phenological period prediction results and coordination strategies. Finally, the controller outputs instructions to drive the actuator to operate according to the updated setpoint, realizing a closed-loop process from growth status perception, decision optimization to control execution. This not only eliminates the disturbances caused by sudden changes in setpoints to the control system, but also ensures the systematicness and coordination of maintenance operations through multi-parameter collaborative optimization, thereby achieving accurate tracking and stable maintenance of the dynamic process of plant growth at the control system level. Attached Figure Description

[0017] Figure 1 This is a flowchart of the landscape plant precision maintenance and control method based on a multi-source sensing and growth model according to the present invention. Figure 2 This is a flowchart illustrating the process of generating a smooth switching trajectory for a set value in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example: Figure 1 This invention presents a method for precise maintenance and control of landscape plants based on a multi-source sensing and growth model, comprising: S1. Continuously collect environmental parameters of landscape plants and physiological state parameters of plants through multi-source sensors; S2. Input the growth environment parameters and plant physiological state parameters into the plant growth model, calculate the optimal environmental settings for the current growth stage, and generate phenological prediction results simultaneously. S3. Based on growth environment parameters and plant physiological state parameters, a causal network of multivariate time series is constructed using a causal discovery algorithm. The temporal evolution entropy of the causal network topology is calculated, and the changing trend of the optimal environmental setpoint is monitored. S4. Identify conflicting relationships between different maintenance objectives based on the causal network topology, and coordinate the priority of objectives by analyzing key causal paths to generate a coordinated maintenance strategy. S5. When the time sequence evolution entropy exceeds the adaptive threshold and the trend of the optimal environmental setpoint is consistent with the direction of change indicated by the phenological period prediction results, a smooth switching trajectory of the setpoint is generated based on the phenological period prediction results and the coordinated maintenance strategy. S6. Based on the set value, smoothly switch the trajectory, adjust the controller's output command, and drive the actuator to perform maintenance work according to the updated set value.

[0020] In step S1, multi-source sensors deployed in the landscape plant growth area continuously collect growth environment parameters and plant physiological state parameters. These multi-source sensors include a temperature sensor for measuring ambient temperature, a humidity sensor for measuring ambient humidity, a light sensor for measuring light intensity, a leaf temperature sensor for measuring leaf surface temperature, and a stem diameter sensor for measuring changes in stem diameter. Specifically, the temperature, humidity, and light sensors collect temperature, humidity, and light intensity data from the growth environment parameters, respectively; the leaf temperature and stem diameter sensors collect leaf temperature and stem diameter change data from the plant physiological state parameters, respectively. The selection of these sensors is based on the typical growth requirements of landscape plants. For example, the temperature sensor uses a digital sensor with a measurement accuracy of ±0.5 degrees Celsius; the humidity sensor uses a capacitive sensor with a measurement accuracy of ±3%; the light sensor uses a photodiode sensor with a measurement range covering 0 lux to 100,000 lux; the leaf temperature sensor uses an infrared thermometer with a measurement accuracy of ±0.2 degrees Celsius; and the stem diameter sensor uses a linear variable differential transformer sensor with a resolution of 0.01 millimeters. The sensor installation locations have been optimized. For example, temperature and humidity sensors are installed at the height of the plant canopy, 1.5 meters above the ground, avoiding direct sunlight and rain. Light sensors are installed in an unobstructed area above the plant canopy, 0.5 meters above the canopy. Leaf temperature sensors use a non-contact measurement method, installed directly onto the leaf surface, 10 to 20 centimeters away from the leaf. Stem diameter sensors use a contact measurement method, fixedly installed at a representative position on the main stem of the plant, 10 centimeters above the ground. All sensor installation locations take into account the spatial distribution characteristics of plant growth; for example, sensor density is increased in densely planted areas and decreased in sparsely planted areas to balance data coverage and cost.

[0021] To achieve synchronous data acquisition from multiple sensors, the system employs a clock synchronization mechanism based on a network time protocol, configuring a unified time reference for all sensor nodes. In practice, the central controller broadcasts a time synchronization signal to each sensor node. Upon receiving the signal, each node simultaneously acquires data according to a preset acquisition cycle. Within each cycle, the temperature sensor collects ambient temperature data, the humidity sensor collects ambient humidity data, the light sensor collects light intensity data, the leaf temperature sensor collects plant leaf temperature data, and the stem diameter sensor collects changes in plant stem diameter. After completing data acquisition, all sensors attach a unified timestamp to each data sample, accurate to the second. The transmission frequency of the time synchronization signal is dynamically adjusted according to the acquisition cycle. For example, when the acquisition cycle is 10 minutes, the time synchronization signal is sent every 5 minutes to ensure that the clock drift error is less than 1 second. The clock synchronization mechanism also includes an error compensation algorithm, such as dynamically adjusting the timestamp offset by calculating signal transmission delay and node processing time to ensure strict alignment of multi-source data in the time dimension.

[0022] To ensure data quality during data acquisition, a data validity verification mechanism was implemented. The verification threshold was set based on a reasonable range of landscape plant growth parameters. This threshold was determined by analyzing historical growth data and plant physiological characteristics. For example, for ambient temperature data collected by temperature sensors, the valid range was set to -10°C to 50°C. This range was determined based on the survival limits of landscape plants in extreme environments and was obtained through long-term observation of plant growth at different temperatures. For ambient humidity data collected by humidity sensors, the valid range was set to 10% to 100%. This range takes into account plant transpiration and the air saturation point, and was set based on experimental data on plant water requirements. For light intensity data collected by the light sensor, the effective range is set to 0 lux to 100,000 lux, covering natural conditions from darkness to strong light, determined with reference to the light response curve of plant photosynthesis. For plant leaf temperature data collected by the leaf temperature sensor, the effective range is set to 0 degrees Celsius to 50 degrees Celsius, calculated based on the temperature limit for plant cell activity and a leaf heat exchange model. For plant stem diameter variation data collected by the stem diameter sensor, the effective range is set to 0 mm to 10 mm, derived from the diurnal diameter variation pattern of plants by statistically analyzing the daily variation amplitude of stems. When the collected data exceeds the corresponding effective range, the system automatically marks it as abnormal data and initiates a re-collection process, with a maximum of 3 re-collections. If the data is still abnormal, it is recorded as invalid data and an alarm is triggered. The data validity verification mechanism also includes data smoothing processing, such as using a moving average method to filter continuously collected data to reduce the impact of random noise.

[0023] The deployment locations of the multi-source sensors were optimized to ensure that the collected data represents the true growth status of the plants. The deployment locations were determined based on plant growth models and spatial statistical analysis, such as identifying hotspots in the plant canopy through cluster analysis and deploying sensors at representative locations. Specifically, temperature and humidity sensors are installed at canopy height, for example, 1.5 meters above the ground, avoiding direct sunlight and rain erosion; light sensors are installed in unobstructed areas above the canopy, for example, 0.5 meters above the canopy, ensuring comprehensive perception of light conditions; leaf temperature sensors use a non-contact measurement method, installed directly on the plant leaf surface, for example, 10 to 20 centimeters away from the leaf, avoiding shadow interference; stem diameter sensors use a contact measurement method, fixedly installed at a representative location on the main stem of the plant, for example, 10 centimeters above the ground, ensuring stable measurement. All sensor installation locations take into account the spatial distribution characteristics of plant growth, for example, increasing sensor density in densely planted areas and decreasing density in sparse areas to balance data coverage and cost. Deployment optimization also includes adjusting sensor orientation, for example, orienting the light sensor due south to maximize light collection efficiency.

[0024] Data acquisition frequency is configured based on plant growth characteristics, which are determined by phenological stages and historical data. For example, a higher acquisition frequency is used during the rapid growth phase, collecting data every 10 minutes; a lower acquisition frequency is used during the slow growth phase, collecting data every 30 minutes. Frequency switching is automatically triggered by analyzing historical growth curves; for example, a rapid growth phase is defined as a daily change in stem diameter exceeding 0.1 mm. The collected growth environment parameters and plant physiological state parameters are transmitted wirelessly to the data storage unit using the IEEE 802.15.4 protocol. Transmission power is adjusted according to distance; for example, 1 milliwatt is used within a 100-meter range. Data encryption and verification mechanisms are employed during transmission. Encryption uses the AES-128 algorithm, and verification uses CRC-32 cyclic redundancy check to ensure data integrity and security. The storage unit categorizes and stores the received data, establishing a structured database containing fields such as timestamp, sensor type, acquired values, and quality indicators. The database uses a relational model; for example, each data table contains a timestamp primary key and sensor value fields, providing a data foundation for subsequent processing and analysis. Data storage also implements a regular backup and cleanup strategy, with a backup cycle of once a week and cleanup rules retaining data from the most recent 365 days to ensure long-term stable operation of the system.

[0025] Through the above implementation methods, strict consistency between growth environment parameters and plant physiological state parameters over time is achieved, ensuring the comparability and synergy of multi-source data used in subsequent processing and analysis. This synchronous acquisition mechanism effectively avoids data deviations caused by asynchronous acquisition times, providing data quality assurance for accurate calculation of plant growth models and reliable construction of causal networks. The entire acquisition process is automated, requiring no manual intervention, and can continuously and stably acquire multi-source sensor data reflecting plant growth status. The data acquisition system also includes self-checking functions, such as periodically performing sensor calibration once a month, using standard reference comparisons to ensure measurement accuracy. System operating status is recorded through logs, such as recording acquisition success rate, abnormal data ratio, and transmission latency, facilitating maintenance and optimization.

[0026] In addition, soil environmental parameters and more comprehensive plant physiological state parameters can be collected to further enhance the completeness of state perception. Soil environmental parameter collection includes soil moisture, soil pH, and soil nutrient content. For example, frequency domain reflectance sensors are used to measure soil volumetric water content, ion-selective electrode sensors are used to measure soil pH, and electrochemical sensors are used to measure the content of nitrogen, phosphorus, and potassium elements in the soil. More comprehensive plant physiological state parameters include leaf water status, root activity, and visual characteristics of pests and diseases. For example, leaf water content sensors are used to measure relative leaf water content using the dielectric constant method, microroot canal imaging systems are used in conjunction with image analysis to assess root activity, and multispectral cameras are used to acquire canopy images to extract spectral features related to pests and diseases. Soil environmental parameters, plant physiological state parameters, and growth environment parameters are collected synchronously, aggregated, and assigned a unified timestamp, collectively forming a more complete multidimensional time-series data foundation.

[0027] In step S2, the growth environment parameters and plant physiological state parameters are input into the plant growth model to calculate the optimal environmental setpoints for the current growth stage, and phenological prediction results are generated simultaneously. The plant growth model calculates the optimal environmental setpoints based on the input growth environment parameters and plant physiological state parameters through a pre-defined growth stage mapping relationship. The growth stage mapping relationship is pre-established according to plant growth characteristics. For example, by analyzing the environmental parameter requirements of landscape plants at different growth stages, the plant growth process is divided into multiple continuous stages, each corresponding to a set of optimal environmental setpoints. The establishment of the growth stage mapping relationship is based on historical observation data and plant physiological knowledge. For example, by long-term monitoring of the correlation between environmental parameters and growth status during the budding, growth, flowering, and fruiting stages, the optimal temperature range, optimal humidity range, and optimal light range for each stage are determined. The calculation of the optimal environmental setpoints is achieved by querying the growth stage mapping relationship. For example, when the input parameters indicate that the plant is in the growth stage, the model outputs the optimal temperature setpoint, optimal humidity setpoint, and optimal light setpoint corresponding to the growth stage. The plant growth model also includes a parameter adjustment mechanism, such as dynamically fine-tuning the optimal environmental setpoints based on real-time collected plant physiological parameters to ensure that the setpoints adapt to the actual growth state of the plant. This parameter adjustment mechanism is based on feedback control principles; for example, when leaf temperature continuously deviates from the expected range, the optimal temperature setpoint is automatically adjusted, with the adjustment range determined by the degree of deviation and the plant's tolerance.

[0028] The phenological stage prediction algorithm establishes a prediction model by analyzing historical growth environment parameters, plant physiological state parameters, and phenological stage observation records. The prediction model is built based on multiple regression analysis, for example, collecting historical data on growth environment parameter sequences and plant physiological state parameter sequences as independent variables, and phenological stage observation records as dependent variables, then fitting the regression equation using the least squares method. The training process of the phenological stage prediction algorithm uses historical growth data, for example, selecting data from the past three years' complete growth cycles as the training set. This data includes parameters such as daily average temperature, cumulative light duration, leaf surface temperature change rate, and stem diameter growth rate, as well as corresponding phenological stage labels. The parameters of the regression equation are determined through iterative optimization, for example, setting the loss function as mean squared error, using gradient descent to update the parameters until convergence, with the convergence condition set as the loss function change rate being less than 0.001. After training, the current growth environment parameters and plant physiological state parameters are input into the prediction model, and the phenological stage state value is calculated using the multiple regression equation. The phenological stage state value is a continuous numerical value, for example, ranging from 0 to 1, representing the degree of progress of the plant from dormancy to maturity. The weights of the independent variables in the multiple regression equation are automatically learned through the training process. For example, the initial weight values ​​are randomly set and updated according to the gradient direction during the iteration process.

[0029] Phenological state values ​​are mapped to preset phenological stages to generate phenological prediction results. The preset phenological stages are based on plant phenological characteristics, such as dividing phenological periods into dormancy, budding, leaf expansion, flowering, fruiting, and leaf fall. The mapping process is achieved by setting phenological state thresholds. For example, a phenological state value below 0.2 corresponds to dormancy, 0.2 to 0.4 corresponds to budding, 0.4 to 0.6 corresponds to leaf expansion, 0.6 to 0.8 corresponds to flowering, 0.8 to 0.9 corresponds to fruiting, and greater than 0.9 corresponds to leaf fall. The phenological state thresholds are obtained based on statistical analysis of historical phenological observation data. For example, the distribution range of phenological state values ​​corresponding to each phenological stage is calculated, and the distribution boundary is taken as the threshold. The distribution boundary is determined by calculating the percentile of the state value for each stage, for example, using the 10th percentile and the 90th percentile as the threshold boundary. The generation of phenological period prediction results also includes uncertainty assessment, such as indicating the reliability of the prediction by calculating the confidence interval of the predicted value. The confidence interval is calculated based on the standard error of the regression model, which is calculated by the ratio of the sum of squared residuals to the degrees of freedom.

[0030] The coordinated operation of the plant growth model and phenological prediction algorithm ensures consistency between the optimal environmental settings and the phenological prediction results. For example, when the phenological prediction indicates that the plant is about to enter the flowering stage, the plant growth model will prioritize adjusting the optimal environmental settings to meet the flowering requirements. The adjustment logic is based on preset association rules between phenological stages and environmental parameters. The model update mechanism is executed periodically, for example, retraining the phenological prediction algorithm monthly to incorporate the latest growth data. During retraining, a sliding window method is used to retain data from the most recent three years. Data preprocessing steps are performed before inputting into the model, such as standardizing growth environment parameters and plant physiological state parameters to eliminate dimensional differences. The standardization method uses min-max normalization, scaling each parameter value to the range of 0 to 1. The normalization parameters are set based on the minimum and maximum values ​​of historical data. Outlier handling uses interpolation to fill missing values; for example, linear interpolation is used to estimate missing values ​​based on data from preceding and following time points, with the interpolation window set to 5 time points before and after. The entire calculation process is implemented on an embedded processor, with the calculation frequency synchronized with the data acquisition frequency, for example, performing model calculations every 10 minutes. The calculation results are stored in a database, providing a basis for maintenance strategy decisions. The database storage format includes fields such as timestamp, model version, optimal environmental settings, phenological period prediction results, and confidence level to ensure data traceability.

[0031] In step S3, based on growth environment parameters and plant physiological state parameters, a causal network of multivariate time series is constructed using a causal discovery algorithm. The temporal evolution entropy of the causal network topology is calculated, and the changing trend of the optimal environmental setpoint is monitored. The causal discovery algorithm processes the time series data of growth environment parameters and plant physiological state parameters, establishes causal relationships between parameters, and constructs a causal network. The causal discovery algorithm employs a method based on conditional independence testing, such as the PC algorithm, to infer causal relationships by analyzing the conditional dependencies between various parameters in the multivariate time series. The algorithm execution process includes three main stages: data preprocessing, conditional independence testing, and directionality determination. In the data preprocessing stage, the time series of growth environment parameters and plant physiological state parameters undergo stationarity testing and detrending processing, for example, using differencing methods to eliminate trend components in the time series, ensuring that the data meets the stability requirements of causal discovery. In the conditional independence testing stage, the conditional dependencies between parameters are assessed by calculating partial correlation coefficients. For example, a significance level threshold of 0.05 is set to determine conditional independence; the significance level threshold is determined based on conventional statistical testing practices. The directionality determination stage utilizes time series information to infer causal direction, such as determining causal direction based on the time series constraint principle that cause precedes effect. Data preprocessing also includes handling missing values, such as using linear interpolation to fill data gaps, with the interpolation window set to 5 time points before and after the data.

[0032] The construction of causal networks is based on the output of causal discovery algorithms. Each growth environment parameter and plant physiological state parameter is represented as a network node, and the determined causal relationships are represented as directed edges. The weights of network edges are quantified using causal strength indices, such as standardized partial correlation coefficients, with weight values ​​ranging from 0 to 1. The degree centrality of nodes in the causal network is calculated by summing the out-degree and in-degree; for example, the degree centrality of a node equals the sum of the number of edges pointing to that node and the number of edges pointing to other nodes. The causal network is stored in an adjacency matrix format, where matrix elements represent the causal strength between nodes, and the matrix dimension equals the number of parameters. The calculation of node degree centrality also includes normalization processing, such as dividing the degree centrality value by the maximum possible degree centrality value to eliminate the influence of differences in network size.

[0033] The temporal evolution entropy is calculated based on the changes in node connectivity in a causal network. The calculation of temporal evolution entropy is based on the time-series rate of change of node degree centrality in the causal network topology. Node degree centrality time series are obtained through a sliding time window, for example, setting the time window length to 24 hours and the sliding step size to 1 hour, continuously acquiring node degree centrality values ​​within multiple time windows. The rate of change of node degree centrality between adjacent time windows is calculated, for example, by subtracting the node degree centrality value of the previous time window from the node degree centrality value of the later time window, and then dividing by the node degree centrality value of the previous time window to obtain the relative rate of change. If the node degree centrality value of the previous time window is 0, the rate of change is set to a specific default value, such as 0, to avoid division by zero errors. The standard deviation is calculated for all node degree centrality change rate sequences to obtain the temporal evolution entropy. The calculation of temporal evolution entropy also includes normalization processing, such as dividing the standard deviation by the number of nodes, to eliminate the influence of network size on the entropy value. The update frequency of the temporal evolution entropy is consistent with the step size of the sliding time window, for example, the latest temporal evolution entropy value is calculated once per hour.

[0034] Simultaneously, the changing trend of the optimal environmental setpoint is monitored by analyzing its numerical changes over time. The changing trend of the optimal environmental setpoint is calculated through numerical differences at consecutive time points. The numerical difference employs a first-order forward differencing method, for example, subtracting the optimal environmental setpoint from the previous time point to obtain the instantaneous change. The determination of the changing trend is based on the analysis of changes at multiple consecutive time points; for example, when the changes at three consecutive time points maintain the same sign, a significant changing trend is considered to exist. The strength of the changing trend is quantified by the moving average of the changes; for example, the arithmetic mean of the changes at the most recent five time points is calculated as the trend strength indicator. The monitoring of changing trends also includes abnormal fluctuation detection; for example, when the change at a single time point exceeds the normal fluctuation range, it is marked as an anomaly. The normal fluctuation range is determined based on the statistical distribution of historical changes; for example, the mean and standard deviation of historical changes are calculated, and the normal fluctuation range is set to the mean plus or minus twice the standard deviation.

[0035] The calculated results of temporal evolution entropy are correlated with the monitoring results of the changing trend of the optimal environmental setpoint. For example, when the temporal evolution entropy value increases significantly and the optimal environmental setpoint shows a clear upward trend, the system is judged to be in a state transition period. The correlation analysis is based on preset correlation rules, such as setting a combination of temporal evolution entropy threshold and changing trend intensity threshold. When both conditions are met simultaneously, a state warning is triggered. The temporal evolution entropy threshold is set based on statistical analysis of historical data, such as calculating the distribution characteristics of temporal evolution entropy values ​​over the past month and taking the 90th percentile as the threshold. The changing trend intensity threshold is set based on the normal variation range of the optimal environmental setpoint, such as determining the threshold range by analyzing the typical variation amplitude of the setpoint during the plant growth stage transition. The threshold update mechanism is executed periodically, such as recalculating the threshold weekly to reflect the latest system state.

[0036] The entire computation process employs a streaming architecture, updating the causal network and temporal evolution entropy upon receiving new data points. The execution frequency of the causal discovery algorithm is configured based on data characteristics, for example, re-executing causal discovery every 6 hours to capture dynamic changes in causal relationships between parameters. Data storage during the computation process utilizes a time-series database, storing data on the causal network structure, node degree centrality sequences, temporal evolution entropy sequences, and trends in optimal environmental setpoints. The system also implements computational quality monitoring, such as periodically evaluating the stability of the causal network and prompting parameter recalibration when network structure changes significantly. All computation results are stored with timestamps, providing comprehensive historical data support for subsequent decision-making. Data storage employs compressed formats to reduce storage space, such as using differential encoding to compress time-series data. The system also includes error recovery mechanisms, such as resuming processing from the most recent checkpoint when computation is interrupted, with checkpoints set at 1-hour intervals.

[0037] In step S4, conflict relationships between different maintenance objectives are identified based on the causal network topology, and the priority of objectives is coordinated by analyzing key causal paths to generate a coordinated maintenance strategy. The conflict relationships are identified based on the connection strength and direction between nodes in the causal network topology. Maintenance objective nodes are determined through a pre-established mapping table; for example, growth promotion objectives are mapped to growth rate nodes, and resource conservation objectives are mapped to resource consumption nodes. Conflict relationship identification is achieved by analyzing the connection patterns between nodes; for example, a conflict is determined when there is a negatively correlated connection path between two maintenance objective nodes. The connection strength is evaluated using edge weights in the causal network, which are derived from the standardized partial correlation coefficient calculated by the causal discovery algorithm, with values ​​ranging from -1 to +1. The conflict determination threshold is set to zero; for example, a conflict is confirmed when at least one edge on the connection path between two objective nodes has a weight less than zero. Conflict relationship identification also includes intensity quantification, such as calculating a conflict intensity index by accumulating the absolute values ​​of the weights of all negatively weighted paths. The standardization of the conflict intensity index is achieved by dividing by the maximum possible conflict intensity value, for example, the maximum possible conflict intensity value is taken as the 95th percentile of the historical conflict intensity values.

[0038] The process involves extracting critical causal paths connecting nodes representing different maintenance goals within a causal network topology. This extraction includes identifying nodes representing different maintenance goals and searching for all paths representing these goals within the causal network topology. A depth-first search algorithm is used for path finding, with a maximum path length set at 5 nodes. This length is determined statistically based on the average path length of the causal network to avoid excessively long, invalid paths. The causal influence strength of a path is calculated by multiplying the connection weights between adjacent nodes. For example, the causal influence strength of a path from the starting node to the target node is equal to the continuous product of the connection weights between all adjacent nodes on the path. The calculation of causal influence strength also includes normalization, such as dividing the product by the path length to eliminate the influence of path length on the strength value. The path with the highest causal influence strength is selected as the critical causal path. For example, the causal influence strength of all paths connecting the same pair of maintenance goal nodes is compared, and the path with the highest strength value is selected as the critical causal path. Critical causal paths are stored in a path sequence format, recording the sequence of nodes traversed by the path and their corresponding causal influence strength values.

[0039] For conflicting relationships, the causal influence strength along each key causal path is analyzed, and maintenance objectives with conflicting relationships are prioritized based on this strength. Prioritization is based on the relative magnitude of causal influence strength; for example, maintenance objectives with greater causal influence strength are assigned higher priority. Prioritization also considers the importance weight of maintenance objectives; for example, an expert evaluation method is used to assign importance weights to each maintenance objective. This expert evaluation method employs a multi-round Delphi method to collect opinions from domain experts, and the importance weights range from 0 to 1. The final priority score is calculated by multiplying the causal influence strength by the importance weight; for example, the priority score equals the causal influence strength of the key causal path multiplied by the importance weight of the corresponding maintenance objective. Maintenance objectives are then ranked according to their priority scores, for example, determining their execution priority in descending order of score. The priority ranking update mechanism is based on system state changes; for example, the priority ranking is recalculated when new conflicting relationships are detected.

[0040] The coordinated maintenance strategy is generated based on priority ranking. For example, when two maintenance goals conflict, the higher-priority goal is prioritized. The coordinated strategy includes specific parameter adjustment schemes; for instance, for the high-priority goal of promoting growth, temperature and light intensity settings are appropriately increased, while for the low-priority goal of conserving resources, humidity settings are correspondingly decreased. The strategy generation also includes conflict resolution rules; for example, when resource consumption exceeds a threshold, the resource allocation ratio for lower-priority goals is automatically reduced. This threshold is set based on the 80th percentile of historical resource usage data. The coordinated maintenance strategy is output as executable instructions, including a list of goal priorities, parameter adjustments, and execution sequence. A strategy verification mechanism ensures the generated maintenance strategy conforms to actual constraints, such as checking if parameter adjustments are within the equipment's allowable range and automatically adjusting to reasonable values ​​when they exceed the range. Strategy verification also includes consistency checks, ensuring no logical contradictions exist between different maintenance strategies.

[0041] The entire process employs an iterative optimization approach, such as reassessing conflict relationships and priority ranking every 24 hours to adapt to changes in system status. Historical strategy storage is used for strategy optimization, such as saving maintenance strategies from the past 30 days and their implementation effects, and improving the priority ranking algorithm by analyzing historical data. All intermediate results, including conflict relationship identification results, critical causal path sets, priority ranking results, and coordinated maintenance strategies, are recorded in the strategy database for subsequent analysis and debugging. The strategy database uses a relational database and includes fields such as timestamps, conflict pairs, critical paths, priority scores, and maintenance strategies to ensure data traceability and analyzability. The data processing also includes exception handling; for example, when a critical causal path cannot be found, a default priority rule is applied, determined based on the fundamental importance of the maintenance objectives.

[0042] Figure 2 A flowchart of the process for generating a smooth switching trajectory for setpoints according to this invention is provided. In step S5, when the temporal evolution entropy exceeds the adaptive threshold and the trend of change in the optimal environmental setpoint is consistent with the direction of change indicated by the phenological period prediction results, a smooth switching trajectory for setpoints is generated based on the phenological period prediction results and the coordinated maintenance strategy. The degree of change in system state is determined by comparing the temporal evolution entropy with the adaptive threshold dynamically adjusted based on historical data. The adaptive threshold is calculated using a sliding window statistical method, for example, using the temporal evolution entropy data of the past 30 days and calculating its 75th percentile as the current adaptive threshold. The selection of this percentile is based on the upper limit of the normal fluctuation range in historical data analysis. The update cycle of the adaptive threshold is set to 24 hours, for example, the latest adaptive threshold is recalculated at midnight every day. The update process includes data validity checks, for example, using a default threshold when the amount of historical data is insufficient. The comparison between temporal evolution entropy and adaptive threshold is performed using a relative comparison method. For example, when the temporal evolution entropy exceeds 1.2 times the adaptive threshold, the degree of change in system state is determined to meet the trigger condition. This multiple factor is determined through sensitivity analysis. For example, the optimal value is selected after testing the impact of different multiples on the accuracy of system state detection.

[0043] The rationality of the growth stage transition is verified by analyzing the consistency between the changing trend of the optimal environmental setpoints and the direction of change indicated by the phenological stage prediction results. The consistency test of the changing trend employs a direction matching algorithm. For example, the changing trend of the optimal environmental setpoints is quantified as a direction indicator, which is determined by calculating the sign of the change in the setpoints at the three most recent time points. The number of time points is determined based on a balance between data sampling frequency and trend stability requirements. The direction of change indicated by the phenological stage prediction results is determined by phenological stage transition rules. For example, when the phenological stage prediction results show a transition from the growth stage to the flowering stage, the direction of change is indicated by an increase in the temperature setpoint and an increase in the light setpoint. The transition rules are predefined based on plant physiology knowledge. Consistency verification is achieved by comparing whether the direction of change indicator matches the direction of change indicated by the phenological stage prediction results. For example, if the signs of the two direction indicators are consistent, the verification is considered successful. The verification process includes uncertainty handling; for example, when the direction of change is unclear, a verification time window is added.

[0044] When both the degree of system state change and the rationality of growth stage transition are met, the target growth stage is determined based on phenological prediction results. For example, if the phenological prediction results indicate that the plant is in the flowering stage, the target growth stage is determined to be the flowering stage. This is combined with the priority of objectives in the coordinated maintenance strategy; for example, the highest priority maintenance objective in the coordinated maintenance strategy is selected as promoting flowering. A smooth transition function is used to generate a smooth switching trajectory of continuously changing setpoints over time. The smooth transition function uses an S-shaped curve function, such as using the logistic function as the basic form of the transition function. The slope parameter of the function is calculated based on the transition time and the magnitude of the setpoint change. The transition time is determined based on the degree of difference between growth stages; for example, the transition time from the current growth stage to the target growth stage is set to 48 hours, a duration determined through experimental data based on the typical duration of plant growth stage transitions.

[0045] The generation of smooth setpoint transition trajectories involves the synchronous coordination of multiple environmental parameters, such as simultaneously generating smooth transition trajectories for temperature, humidity, and light settings. The transition trajectory for each parameter is calculated based on its target setpoint, which is obtained by querying an optimal environmental setpoint table for the target growth stage. The parameters of the smooth transition function are adjusted according to the plant species; for example, a faster transition speed is set for flowering plants, and a slower transition speed is set for foliage plants. The adjustment rules are pre-set based on plant growth characteristics. Trajectory generation also considers equipment response characteristics; for example, for temperature control equipment with a slow response, the transition time is appropriately extended to ensure stable equipment operation. The amount of extension is determined through equipment response testing.

[0046] The setpoint smoothing transition trajectory is discretized using a fixed time interval, for example, a 48-hour transition trajectory is discretized into 96 time points, each with a 30-minute time interval, determined based on the control system's execution cycle. The setpoint at each time point is calculated using a smoothing transition function, such as using a logistic function to calculate the proportion of the setpoint at a specific time point, followed by linear interpolation based on the initial and target setpoints. Trajectory verification ensures the generated setpoint smoothing transition trajectory conforms to the equipment's operating range; for example, it checks whether the setpoints at all time points are between the equipment's allowed minimum and maximum values, automatically adjusting to boundary values ​​when they exceed the range. Trajectory storage uses a time-series format, with each time point containing a timestamp and a corresponding array of setpoints.

[0047] The entire generation process employs an exception handling mechanism. For example, if a smooth transition trajectory with valid set values ​​cannot be generated, a default step transition method is used, selected based on historical successful transition cases. Trajectory optimization is based on historical execution results; for instance, plant response data is recorded after each trajectory execution, and the parameters of the smooth transition function are optimized using machine learning methods, with an optimization cycle of once a week. All intermediate data, including adaptive thresholds, consistency verification results, target growth stages, and smooth transition function parameters, are recorded in the generation log to ensure process traceability. The generation log contains fields such as timestamps, input parameters, processing steps, and output results, and is stored in a dedicated trajectory generation database. Database maintenance includes periodic cleanup, such as retaining the generation logs for the most recent 90 days for analysis.

[0048] In step S6, based on the setpoint smoothing switching trajectory, the controller's output command is adjusted to drive the actuator to perform maintenance operations according to the updated setpoints. The setpoint smoothing switching trajectory is discretized into a sequence of target setpoints for multiple control cycles. The length of the control cycle is determined based on the response characteristics of the actuator. For example, for ventilation equipment with a fast response, a shorter control cycle, such as 30 seconds, is set based on the time it takes to reach a steady state; for irrigation equipment with a slow response, a longer control cycle, such as 5 minutes, is set based on the water flow propagation delay. The discretization process uses an equally spaced sampling method, for example, extracting the corresponding target setpoints from the setpoint smoothing switching trajectory according to the control cycle length to form a temporally continuous sequence of target setpoints. The target setpoint sequence is stored using a circular buffer structure, for example, setting the buffer length to 100 control cycles. This length is determined based on a balance between the control system's memory capacity and real-time requirements. When the buffer is full, the oldest data is automatically overwritten.

[0049] Based on the deviation between the target setpoint sequence and the current actual measured value, the controller's output command is calculated using a control algorithm. The control algorithm employs a proportional-integral-derivative (PID) control algorithm, where the tuning of the proportional gain, integral time, and derivative time parameters is based on the dynamic characteristics of the actuator. For example, specific parameter values ​​are determined during the field commissioning phase using the Ziegler-Nichols method. The tuning process involves gradually increasing the proportional gain until the system exhibits sustained oscillation, recording the critical gain and oscillation period, and then calculating the proportional gain, integral time, and derivative time according to the formula. Deviation calculation uses real-time measurement methods, such as acquiring the actual measured values ​​of the current environmental parameters at the beginning of each control cycle and calculating the difference between these values ​​and the target setpoint at the corresponding time. The control algorithm's output command is calculated using a discretized PID formula; for example, the output command equals the proportional gain multiplied by the deviation, plus the integral of the integral of the deviation, plus the derivative of the derivative multiplied by the derivative of the deviation. Output command limiting ensures that the command value remains within the actuator's operating range; for example, the output command is limited to between 0 and 100%, corresponding to the actuator's fully closed to fully open state.

[0050] The controller's output commands drive the actuators to track the target setpoint sequence of environmental parameters. The actuators include irrigation equipment, shading equipment, and ventilation equipment, corresponding to the regulation of environmental parameters such as moisture, light intensity, and temperature, respectively. Irrigation equipment is controlled using pulse width modulation (PWM), for example, adjusting the duty cycle of the solenoid valve's opening time based on the output command value to achieve precise water volume control. Shading equipment is controlled using position servo control, for example, controlling the extent of the shading net's deployment based on the output command value to adjust light intensity. Ventilation equipment is controlled using variable frequency speed regulation, for example, adjusting the fan speed based on the output command value to control airflow. Each actuator is equipped with a local feedback sensor; for example, irrigation equipment is equipped with a flow sensor, shading equipment with a position sensor, and ventilation equipment with a speed sensor, forming a closed-loop control system.

[0051] Modeling the response characteristics of actuators is used to optimize control performance. For example, a transfer function model of ventilation equipment is established, and model predictive control improves the accuracy of temperature regulation. Equipment status monitoring detects the operating status of actuators in real time, such as monitoring abnormal fluctuations in motor current. When an anomaly is detected, it automatically switches to standby equipment or triggers an alarm. The sampling time synchronization of the control system uses a hardware clock, such as a real-time operating system, to ensure accurate timing of each control cycle, with clock synchronization accuracy reaching the millisecond level. The control log records key data for each control cycle, such as timestamps, target setpoints, actual measured values, output command values, and equipment status. The log data is retained for 90 days.

[0052] The self-tuning function of the control parameters automatically adjusts according to the system's operating status. For example, when a persistently large tracking error of environmental parameters is detected, the proportional, integral, and derivative parameters are automatically retuned. The tuning process employs a gradual method; for instance, the proportional coefficient is first adjusted until the system exhibits critical oscillation, and then the integral and derivative times are calculated based on the oscillation period. A safety protection mechanism ensures reliable operation of the control system. For example, a safe range for environmental parameters is set, and the system immediately switches to a safe mode when the measured value exceeds this range. The safe range is set based on the plant's tolerance limits; for example, the safe temperature range is set from 5 degrees Celsius to 35 degrees Celsius, a range determined through plant physiological experiments.

[0053] The control system uses industrial Ethernet protocols for communication, such as Modbus TCP, to exchange data with actuators. Communication failure handling includes multiple safeguards; for example, in the event of a network interruption, the last valid output command value is used to maintain equipment operation, while local buffer control is activated. Energy optimization strategies are implemented during control, such as selecting the actuator combination with the lowest energy consumption while meeting environmental parameter control requirements. System performance evaluation is performed periodically, such as calculating the average tracking error and control stability index of environmental parameters monthly, and optimizing the control strategy based on the evaluation results.

[0054] The entire control process is initiated by a smooth setpoint transition trajectory. For example, when a new setpoint transition trajectory is detected, the control system is automatically initialized and execution begins. The control process is terminated under two conditions: normal completion and abnormal interruption. For example, control stops when the target setpoint sequence is completed or a major equipment failure is detected. The control results are evaluated by comparing the actual environmental parameter curves with the target setpoint sequence. For example, the root mean square error (RMSE) is calculated as a control quality indicator, which is used to improve subsequent control algorithms. An anomaly handling mechanism covers sensor failure scenarios. For instance, if no valid measurement value can be obtained for three consecutive control cycles, the system automatically switches to model-based predictive control mode, using historical data to predict the current environmental parameter values.

[0055] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0056] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0057] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0058] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0059] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0060] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0061] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0062] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0063] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0064] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for precise maintenance and control of landscape plants based on multi-source sensing and growth models, characterized in that, include: S1. Continuously collect environmental parameters of landscape plants and physiological state parameters of plants through multi-source sensors; S2. Input the growth environment parameters and plant physiological state parameters into the plant growth model, calculate the optimal environmental settings for the current growth stage, and generate phenological prediction results simultaneously. S3. Based on growth environment parameters and plant physiological state parameters, a causal network of multivariate time series is constructed using a causal discovery algorithm. The temporal evolution entropy of the causal network topology is calculated, and the changing trend of the optimal environmental setpoint is monitored. S4. Identify conflicting relationships between different maintenance objectives based on the causal network topology, and coordinate the priority of objectives by analyzing key causal paths to generate a coordinated maintenance strategy. S5. When the time sequence evolution entropy exceeds the adaptive threshold and the trend of the optimal environmental setpoint is consistent with the direction of change indicated by the phenological period prediction results, a smooth switching trajectory of the setpoint is generated based on the phenological period prediction results and the coordinated maintenance strategy. S6. Based on the set value, smoothly switch the trajectory, adjust the controller's output command, and drive the actuator to perform maintenance work according to the updated set value.

2. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 1, characterized in that, In S1, the multi-source sensors include a temperature sensor, a humidity sensor, a light sensor, a leaf surface temperature sensor, and a stem diameter sensor; The growth environment parameters were collected by temperature, humidity and light sensors; Plant physiological parameters were collected by leaf surface temperature sensors and stem diameter sensors; During the data collection process, the multi-source sensors use a unified time reference to collect data synchronously, ensuring that the growth environment parameters and plant physiological state parameters remain consistent over time.

3. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 1, characterized in that, In S2, the plant growth model calculates the optimal environmental settings based on growth environment parameters and plant physiological state parameters through a preset growth stage mapping relationship. And generate phenological prediction results using a phenological prediction algorithm; The growth stage mapping relationship is pre-established based on the plant growth characteristics, and the phenological period prediction algorithm is trained based on historical growth data.

4. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 3, characterized in that, The phenological period prediction algorithm establishes a prediction model by analyzing historical growth environment parameters and plant physiological state parameters with phenological period observation records; inputs current growth environment parameters and plant physiological state parameters into the prediction model, calculates phenological period state values ​​through multiple regression equations; maps phenological period state values ​​to preset phenological period stages, and generates phenological period prediction results.

5. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 1, characterized in that, In S3, the causal discovery algorithm processes time-series data of growth environment parameters and plant physiological state parameters, establishes causal relationships between parameters, and constructs a causal network. Calculate the temporal evolution entropy based on the changes in node connectivity in a causal network; Meanwhile, the trend of change is monitored by analyzing the numerical changes of the optimal environmental setpoint over time. The calculation of temporal evolution entropy is based on the time series change rate of the node centrality index in the causal network topology. The change trend of the optimal environmental setting value is obtained by numerical difference calculation at its continuous time points.

6. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 5, characterized in that, The node centrality index is node degree centrality; the node degree centrality time series is obtained by using a sliding time window, and the rate of change of node degree centrality between adjacent time windows is calculated; the standard deviation of all node degree centrality change rate series is calculated to obtain the temporal evolution entropy.

7. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 1, characterized in that, In S4, conflict relationships are identified based on the connection strength and direction between nodes in the causal network topology. Extract the key causal paths connecting nodes with different maintenance objectives in the causal network topology; For conflicting relationships, the intensity of causal influence on each key causal path is analyzed, and the maintenance objectives with conflicting relationships are prioritized according to the intensity of causal influence on the key causal paths, thereby generating a coordinated maintenance strategy.

8. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 7, characterized in that, The extraction of critical causal paths includes: identifying nodes representing different maintenance goals; finding all paths representing nodes representing different maintenance goals in the causal network topology; calculating the causal influence strength of the path based on the product of the connection weights between adjacent nodes on the path; and selecting the path with the greatest causal influence strength as the critical causal path.

9. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 1, characterized in that, In S5, the degree of system state change is judged by comparing the temporal evolution entropy with the adaptive threshold dynamically adjusted based on historical data. At the same time, the rationality of the growth stage transition is verified by analyzing the consistency between the changing trend of the optimal environmental setpoint and the changing direction indicated by the phenological period prediction results. When both the degree of change in system state and the rationality of the transition of growth stage are met, the target growth stage is determined based on the phenological period prediction results. Combined with the target priority in the coordinated maintenance strategy, a smooth transition function is used to generate a smooth switching trajectory of the set value that changes continuously in the time dimension.

10. The method for precise maintenance and control of landscape plants based on multi-source sensing and growth model according to claim 1, characterized in that, In S6, the setpoint smooth switching trajectory is discretized into a sequence of target setpoints for multiple control cycles; Based on the deviation between the target setpoint sequence and the current actual measured value, the controller's output command is calculated using a control algorithm. The controller's output commands drive the actuator to perform actions, causing environmental parameters to track the target setpoint sequence; The actuators include irrigation equipment, shading equipment, and ventilation equipment, which respectively regulate environmental parameters such as moisture, light, and temperature.

Images