A crop cultivation growth environment intelligent monitoring method

Abstract

The application discloses a crop cultivation growth environment intelligent monitoring method, and relates to the technical field of data processing. The method comprises the following steps: receiving multi-source environment data of a crop growth environment, and dividing the multi-source environment data into a plurality of candidate data windows of different time scales; inputting each candidate data window into a target transpiration prediction model to obtain a transpiration prediction value corresponding to each candidate data window; determining a current window score of each candidate data window based on historical prediction data of each candidate data window and a change rate of an environmental factor; selecting a target window from the plurality of candidate data windows according to the current window score, and performing environmental regulation based on the transpiration prediction value corresponding to the target window. The application can solve the problem that the existing fixed window prediction method cannot adapt to the rhythm of dynamic environmental changes, and realize more accurate crop transpiration demand prediction.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to an intelligent monitoring method for crop cultivation and growth environment. Background Technology

[0002] Transpiration, as a core physiological process of crop water consumption, requires accurate prediction for achieving on-demand irrigation and water-saving efficiency. Currently, when intelligently regulating the greenhouse environment for crop growth based on crop transpiration requirements, environmental data is collected in real time by various sensors deployed within the greenhouse, including those for temperature, humidity, light, and carbon dioxide. Based on mature evapotranspiration models such as FAO, historical environmental data over a fixed time window is used to predict future crop transpiration requirements. This prediction, combined with preset control thresholds, triggers automated control devices such as fans and drip irrigation systems.

[0003] However, this control method cannot adapt to the dynamic changes in the greenhouse environment, leading to inaccurate predictions of crop transpiration. Summary of the Invention

[0004] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide an intelligent monitoring method for crop cultivation and growth environment, which can solve the technical problems that the existing fixed time window transpiration prediction method cannot adapt to the dynamic changes in the greenhouse environment, has a lagging response when the environment changes rapidly, and has reduced prediction accuracy due to noise disturbances introduced by the short window during stable periods.

[0005] This application provides a method for intelligent monitoring of crop cultivation and growth environment, the method comprising: Receive multi-source environmental data on crop growth environment, and divide the multi-source environmental data into multiple candidate data windows with different time scales; Each of the candidate data windows is input into the target transpiration prediction model to obtain the transpiration prediction value corresponding to each of the candidate data windows; The current window score of each candidate data window is determined based on the historical prediction data of each candidate data window and the rate of change of environmental factors. The rate of change of environmental factors is used to dynamically adjust the weight ratio of the historical prediction data in the current window score. Based on the current window score, a target window is selected from multiple candidate data windows, and environmental regulation is carried out based on the evapotranspiration prediction value corresponding to the target window.

[0006] Compared with existing technologies, this scheme divides multi-source environmental data of crop growth into multiple candidate data windows at different time scales. These windows are then input into the target transpiration prediction model to obtain corresponding transpiration prediction values. The current window score is determined based on historical prediction data and the rate of change of environmental factors for each candidate data window. The weight of historical prediction data in the current window score is dynamically adjusted using the rate of change of environmental factors. Finally, the target window and its transpiration prediction value are selected for environmental regulation based on the current window score. This scheme constructs a multi-time-scale candidate data window structure, enabling different windows to capture the temporal differences in environmental data, avoiding the inherent defects of a single fixed window in terms of time sensitivity or data sufficiency. Furthermore, the dynamic adjustment of the weight of historical prediction data in the window score using the rate of change of environmental factors ensures that when the rate of change of environmental factors is high, the weight of historical prediction data is effectively reduced, thus weakening the interference of outdated information on current decisions. When the rate of change of environmental factors is low, historical prediction data retains a higher reference weight to maintain the stability of the prediction results. Therefore, the current window score can adaptively reflect the comprehensive reliability of each candidate window under real-time environmental changes, and thus the transpiration prediction value corresponding to the selected target window achieves an optimized balance between accuracy and response speed, objectively improving the accuracy of transpiration prediction and adaptability to environmental fluctuations, and providing a more accurate and reliable decision-making basis for crop cultivation and growth environment regulation. Attached Figure Description

[0007] Figure 1 A flowchart of an intelligent monitoring method for crop cultivation and growth environment provided in this application; Figure 2 A flowchart illustrating the steps of an intelligent monitoring method for crop cultivation and growth environment provided in this application; Figure 3 A flowchart illustrating the steps of an intelligent monitoring method for crop cultivation and growth environment provided in this application; Figure 4 A flowchart illustrating the steps of an intelligent monitoring method for crop cultivation and growth environment provided in this application. Detailed Implementation

[0008] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0009] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0010] In existing technologies, intelligent monitoring of crop cultivation environments typically employs transpiration prediction methods with fixed time windows. Specifically, the system collects environmental data in real time using various sensors deployed within the greenhouse, including those for temperature, humidity, light, and carbon dioxide. Based on a mature evapotranspiration model, it organizes historical environmental data using fixed-length time windows to predict future crop transpiration demands. Under stable weather conditions and strong diurnal regularity, this method, with its intuitive architecture, simple calculations, and stable response, can meet the basic needs of daily production management and has become a mature paradigm for environmental monitoring and transpiration prediction in facility agriculture.

[0011] However, greenhouses, as semi-closed dynamic systems, exhibit significant time-varying and spatial heterogeneity in their internal environment. Different weather patterns (such as sunny and cloudy days) bring about varying rates of environmental change, and the alternation of diurnal rhythms also generates periodic environmental fluctuations. Existing fixed-window methods, due to their fixed timescale, cannot perceive the inherent rhythm of environmental changes: when the environment changes rapidly, long windows retain too much outdated information, leading to a lag in prediction response; when the environment is stable, short windows introduce unnecessary disturbances by excessively amplifying noise. This static window mechanism struggles to maintain stable prediction accuracy under complex and variable environmental conditions, thus affecting the accuracy of irrigation decisions and water and fertilizer use efficiency.

[0012] To address the aforementioned technical issues, this application constructs a multi-window parallel prediction and real-time scoring mechanism. Continuously collected multi-source environmental data is organized into candidate data windows according to multiple preset candidate time scales, and each window is independently input into the transpiration prediction model to obtain multiple prediction results. Based on this, the rate of change of environmental factors is introduced as a dynamic adjustment parameter, and an adaptive forgetting factor is constructed. An exponentially weighted moving average is applied to the historical prediction error of each window to generate a real-time score reflecting the window's performance under the current environment. Furthermore, for sudden environmental changes caused by transient events such as opening doors and windows, an event impact coefficient and window sensitivity function are constructed to temporarily correct the scores of affected windows. By comparing the real-time scores of each window, the system dynamically selects the transpiration prediction value corresponding to the optimal window as the final monitoring result and generates control commands such as irrigation and ventilation accordingly. This achieves a leap from static windows to dynamic adaptation, significantly improving the prediction accuracy and control reliability of crop transpiration demand under complex and variable environments.

[0013] Below, in conjunction with Figure 1 The intelligent monitoring method for crop cultivation and growth environment provided in this application is described, and the method includes the following steps: Step S20: Receive multi-source environmental data on crop growth environment and divide the multi-source environmental data into multiple candidate data windows with different time scales; Multi-source environmental data refers to a set of multi-dimensional parameters collected by various sensors deployed in the crop growth environment (such as greenhouses and open fields) to describe the state of the crop microenvironment. This set of multi-dimensional parameters typically includes, but is not limited to, parameters such as air temperature, relative humidity, soil volumetric water content, photosynthetically active radiation intensity, carbon dioxide concentration, and wind speed near the canopy.

[0014] In this application, multi-source environmental data serves as the data entry point for the entire monitoring method. Its core value lies in providing a comprehensive and complementary description of the environmental state for subsequent prediction models. A single environmental factor cannot accurately characterize the complex environmental coupling effects affecting crop stomatal conductance and transpiration, while the fusion of multi-source environmental data can significantly improve the model's predictive ability and generalization performance for changes in transpiration rate.

[0015] In practical deployments, computer equipment can connect to a multi-type sensor network via wired connections such as RS-485 bus or wireless connections such as LoRa and ZigBee. For example, a digital temperature and humidity sensor and a quantum optical sensor can be installed 20 centimeters above the crop canopy; frequency domain reflectance soil moisture sensors can be inserted at depths of 10 centimeters and 20 centimeters in the main root zone; and a three-cup anemometer can be installed 50 centimeters above the ground between rows. All sensors collect data at a synchronized sampling frequency, such as once every 30 seconds or once per minute. After the data acquisition unit performs analog-to-digital conversion, timestamp marking, and error verification, the data is uniformly aggregated in the real-time database of the central processing unit of the computer equipment.

[0016] Building upon this foundation, this step does not directly use the raw multi-source environmental data for prediction. Instead, it performs a crucial data processing operation: the continuously flowing real-time data stream is segmented and reorganized according to multiple preset time scales to form several candidate data windows. Each candidate data window contains all multi-source environmental data from all sampling times within its corresponding time span. This feature constructs a multi-scale time-series view set, the technical significance of which lies in the significant temporal scale differences in the response of crop transpiration to different environmental factors. The physiological response of stomata to changes in light intensity is typically completed within seconds to minutes, while the response to the gradual depletion of soil moisture may be delayed for tens of minutes or even hours. By simultaneously constructing multiple windows at different time scales, computer equipment can capture these dynamic features spanning different time dimensions in parallel, providing the necessary structural foundation for subsequent adaptive window selection mechanisms.

[0017] In practice, the central processing unit (CPU) of the computer maintains a fixed-length circular queue buffer, continuously storing all raw sampled data from the past four hours. When a prediction cycle is triggered, the receiving and partitioning module within the computer can, based on a preset list of window lengths (e.g., 10 minutes, 30 minutes, 60 minutes, 120 minutes), trace back from the end of the circular queue (the latest time) and extract data sequences of corresponding time lengths. Each sequence is independently organized into a tensor of shape equal to the time step multiplied by the feature dimension, where the time step is determined by the ratio of the sampling frequency to the window length, and the feature dimension represents the number of different types of multi-source environmental data. These tensors are independent of each other and together constitute the set of candidate data windows input to the subsequent prediction model.

[0018] Understandably, in order to eliminate the slight deviations in sampling time between different sensors, the multi-source environmental data can be timestamped before dividing the window to ensure strict correspondence between environmental data at the same time.

[0019] Optionally, after obtaining multi-source environmental data, this application will also perform preprocessing operations on the multi-source environmental data, wherein the preprocessing includes outlier removal, missing value completion, and normalization of the multi-source environmental data.

[0020] It should be noted that outlier removal refers to the process of identifying and removing distorted data points in multi-source environmental data caused by sensor malfunctions, transmission interference, or sudden drift. This step receives the original acquired environmental parameter sequence, processes any abrupt changes or outliers, and outputs a valid multi-source environmental data sequence after outlier removal. In practice, statistical methods such as the 3σ principle or interquartile range method can be used to detect outlier data; that is, data points exceeding a set confidence interval are identified as outliers and removed.

[0021] Missing value imputation refers to the process of estimating and filling in missing values ​​in multi-source environmental data when there are gaps caused by outlier removal or sensor communication interruptions. This involves using existing data to estimate and fill in these gaps. For example, interpolation algorithms can be used to calculate reasonable values ​​for the missing locations, outputting continuous and complete time-series data. Common implementation methods include linear interpolation, Lagrange interpolation, or imputation based on the mean of preceding and following data.

[0022] Normalization refers to the process of scaling environmental data of different dimensions or orders of magnitude according to a specific ratio, mapping them to a unified specific interval (such as the [0, 1] interval). For example, mathematical transformations can be used to eliminate dimensional differences and output dimensionless standardized environmental data. Specific implementation modes include min-max normalization or Z-score normalization.

[0023] This application introduces preprocessing steps—outlier removal, missing value completion, and normalization—before inputting multi-source environmental data into the target transpiration prediction model, forming a complete data cleaning and standardization mechanism. Outlier removal eliminates noise interference, missing value completion ensures data continuity, and normalization eliminates dimensional differences. Based on the synergistic effect of these preprocessing steps, high-quality, standardized data input can be provided for subsequent candidate data window segmentation and transpiration prediction modeling. This effectively solves the technical problem of low prediction accuracy caused by inconsistent raw data quality, significantly improving the stability and reliability of intelligent monitoring methods for crop cultivation and growth environments.

[0024] Step S30: Input each candidate data window into the target transpiration prediction model to obtain the transpiration prediction value corresponding to each candidate data window; The target transpiration prediction model can be a pre-constructed mathematical calculation model used to calculate the future transpiration demand of crops based on historical environmental data sequences. This model can employ the well-known FAO evapotranspiration model, whose input parameters include environmental factors such as air temperature, relative humidity, net radiation, wind speed, and soil heat flux, and whose output is the predicted transpiration value. Alternatively, it can use a simplified transpiration model with calibrated parameters or a machine learning regression model trained on historical data. The target transpiration prediction model plays a core computational role in the entire technical solution. Its technical significance lies in transforming raw, multi-source environmental data into transpiration demand indicators with clear physiological significance, providing a quantitative basis for subsequent irrigation decisions. By inputting multiple candidate data windows into the same model, multiple independent prediction results based on different historical information lengths can be obtained. The degree of difference between these results inherently implies information about environmental dynamics.

[0025] For example, the computer device may have a locally calibrated FAO model pre-loaded within the central processing unit. The model's parameters, such as aerodynamic drag and surface drag, have been calibrated according to local crop varieties, planting densities, and greenhouse structural characteristics. After step 20 generates the candidate data window set, the input acquisition module loads the tensor data for each window sequentially, formats the environmental parameters at each time point within the window according to model requirements, calculates the instantaneous evapotranspiration value at each time point, integrates the calculation results for all time points within the candidate data window, and finally outputs the predicted evapotranspiration value for the future time point corresponding to that candidate data window. This process can be executed in parallel for all candidate data windows, generating a set of predicted values ​​equal in number to the number of candidate data windows, with each predicted value accompanied by a length identifier for the corresponding window.

[0026] Step S40: Determine the current window score of each candidate data window based on the historical prediction data of each candidate data window and the environmental factor change rate. The environmental factor change rate is used to dynamically adjust the weight ratio of historical prediction data in the current window score. Historical prediction data can refer to historical errors, historical window scores, etc. generated by each candidate data window over a past period of time.

[0027] The rate of change of environmental factors refers to the magnitude of change of multiple different types of environmental data within a unit of time, that is, the magnitude of change of environmental data corresponding to different types of environmental factors within a unit of time. If the environmental factor is air temperature, then the rate of change of environmental factors can be the absolute value of the first difference of air temperature; if the environmental factor is relative humidity, then the rate of change of environmental factors can be the absolute value of the first difference of relative humidity; if the environmental factor is carbon dioxide concentration, then the rate of change of environmental factors can be the absolute value of the first difference of carbon dioxide concentration; if the environmental factor is photosynthetically active radiation, then the rate of change of environmental factors can be the absolute value of the first difference of photosynthetically active radiation, and so on. The rate of change of environmental factors can quantify the severity of current environmental changes and provide a basis for the dynamic adjustment of candidate data window scoring. When the environment changes rapidly, long windows often have larger prediction errors because they contain too much outdated information. In this case, the scoring mechanism should give higher weight to recent prediction errors. When the environment is stable, short windows are easily affected by noise. In this case, the scoring mechanism should make full use of historical information for smoothing to obtain more accurate monitoring results.

[0028] In practical implementation, the current window score can be calculated using, for example, an exponentially weighted moving average algorithm, with a forgetting factor used to control the retention of historical scores. A higher rate of change in environmental factors indicates more drastic environmental fluctuations; in this case, the forgetting factor should be reduced to decrease the weight of historical data, allowing the current window score to respond more quickly to current prediction errors. Conversely, a lower rate of change in environmental factors indicates a more stable environment; in this case, the forgetting factor should be increased to increase the weight of historical data, using long-term data to smooth out noise. Thus, each candidate data window receives a real-time current window score, and the score directly reflects the predictive performance of that candidate data window under the current rate of environmental change.

[0029] For example, this application may first determine the environmental factor change rate sequence of each environmental factor within each candidate data window and the change rate sequence of the corresponding historical period or similar period. Then, it may use cosine similarity or Euclidean distance to calculate the matching degree between the current change rate and the historical change rate, and use this as the dynamic weight of the historical data. Finally, it may use historical prediction data (such as the average or trend of evapotranspiration in the same historical period) as a benchmark and combine it with the dynamic weight to score.

[0030] Alternatively, this application may also be based on Figure 2 The steps shown are to obtain the current window rating: Step S201: Calculate the prediction error of each candidate data window based on the historical prediction data of each candidate data window; The prediction error refers to the difference between the predicted transpiration values ​​output by each candidate data window and the actual transpiration values ​​at the same moment. The actual transpiration values ​​can be obtained in real time through a high-precision lysimeter deployed in the cultivation medium, or they can be calculated posteriorly based on actual meteorological data using the FAO model; this application does not limit the specific method used. The prediction error transforms the abstract quality of a window's performance into a quantifiable numerical value, providing the initial error input for subsequent scoring updates. A smaller prediction error indicates a more accurate prediction by the candidate data window at the current moment; a larger prediction error indicates a greater deviation from reality.

[0031] In practical implementation, existing error calculation formulas such as root mean square error (RMSE) and mean absolute error (MAE) can be used to process historical prediction data. For example, for each candidate data window, its predicted evapotranspiration value at a historical moment is obtained, and combined with the actual measured evapotranspiration value at that moment (e.g., measured by a lysimeter or calculated using a standard model), the absolute or squared difference between the two is calculated as the instantaneous prediction error of that candidate data window, used for subsequent score updates. This application establishes an objective and quantitative window performance evaluation index by accurately calculating the historical prediction error of each candidate data window, laying the foundation for dynamically adjusting the scoring weights based on the rate of change of environmental factors, thereby solving the problem of the lack of quantitative means for window prediction accuracy in existing technologies.

[0032] For example, this application can call the stored historical prediction logs for candidate data windows of different lengths such as 30 minutes, 1 hour, 2 hours and 4 hours respectively, obtain the prediction values ​​of all sampling points of each candidate data window in the past 1 hour, use the actual evapotranspiration value of the same period calculated by the FAO-56 standard model as the benchmark, calculate the average absolute error between the prediction value of each window and the benchmark value, and thus obtain the error basis data of each candidate data window at the current time.

[0033] Step S202: Determine the change rates of multiple environmental factors based on multi-source environmental data, and construct a comprehensive environmental change rate index based on the change rates of each environmental factor. Among them, the rate of change of environmental factors can be a parameter characterizing the severity of the fluctuation of each environmental factor over time; the comprehensive environmental change rate index can be a quantitative value of overall environmental fluctuation after integrating the rate of change of multiple environmental factors. This step receives real-time multi-source environmental data, then determines the instantaneous rate of change of each environmental factor according to the environmental data sequence corresponding to each environmental factor, and performs weighted or summed fusion to output an index value that can comprehensively reflect the current dynamic changes of the environment, that is, the comprehensive environmental change rate index.

[0034] In an optional embodiment, this application can obtain the comprehensive environmental change rate index in the following manner: At the start of each prediction period, environmental data of various types between the current and previous sampling times are read from the real-time database of the computer equipment. Then, the first-order absolute difference values ​​of air temperature, relative humidity, carbon dioxide concentration, and photosynthetically active radiation are calculated separately. These four individual rates of change may have different dimensions. Therefore, each rate of change can be dimensionless by dividing it by its historical maximum variation, mapping the values ​​to the range of 0 to 1. Subsequently, the processed rates of change can be summed to obtain a comprehensive environmental change rate index, as shown in the following formula:

[0035] in: As a comprehensive environmental change rate index, , , , These are the rates of change of air temperature, the rates of change of relative humidity, and The rate of change and the rate of change of light intensity are theoretically applicable to all environmental data. These four are the most common data and are only used as preferred options without limitation.

[0036] In another optional embodiment, after dimensionless processing of each rate of change, a preset weighting coefficient can be assigned to each rate of change based on the contribution of each environmental factor to crop transpiration. For example, the rate of change in light intensity can be assigned a weight of 0.35, the rate of change in air temperature a weight of 0.30, the rate of change in relative humidity a weight of 0.25, and the rate of change in carbon dioxide concentration a weight of 0.10. Finally, the weighted sum of each individual rate of change is obtained to obtain a comprehensive environmental change rate index, as shown in the following formula:

[0037] in: As a comprehensive environmental change rate index, , , , These are the rates of change of air temperature, the rates of change of relative humidity, and The rate of change and the rate of change of light intensity are theoretically applicable to all environmental data. These four are the most common data and are only used as preferred options without limitation. These are the rates of change of air temperature, the rates of change of relative humidity, and Weights corresponding to the rate of change and the rate of change of light intensity.

[0038] Step S203: Dynamically adjust the forgetting factor based on the comprehensive environmental change rate index; The forgetting factor, in this context, refers to a coefficient used to control the weighting of historical data in an exponentially weighted moving average calculation. This application synchronizes the score update rate with the pace of environmental change by setting a forgetting factor. When environmental changes are drastic, past prediction errors need to be quickly forgotten so that the score can respond rapidly to new environmental changes; in this case, the forgetting factor should be relatively small. When the environment is relatively stable, historical information needs to be fully utilized to smooth out instantaneous noise and avoid unnecessary oscillations in the score due to random fluctuations; in this case, the forgetting factor should be relatively large. This application achieves adaptive synchronization between the score update rate and the speed of environmental change by dynamically adjusting the forgetting factor, avoiding prediction lag or noise amplification caused by a fixed forgetting factor, thereby improving the timeliness and robustness of the window scoring.

[0039] In an optional embodiment, this application may first preset the value range of the forgetting factor, including a first forgetting factor and a second forgetting factor. The first forgetting factor may correspond to the value when environmental changes are most drastic, and is set to 0.8 according to numerical stability requirements, for example; the second forgetting factor may correspond to the value when the environment is most stable, and is set to 0.99 according to historical data retention requirements, for example. Then, a sensitivity coefficient is preset to control the decay rate of the forgetting factor as the comprehensive environmental change rate index changes. When the comprehensive environmental change rate index output from the above steps is input, it may be substituted into a preset nonlinear mapping function for calculation. This function is configured such that the forgetting factor monotonically decreases as the comprehensive environmental change rate index increases, and its value is limited to between the first forgetting factor and the second forgetting factor. It should be noted that if the comprehensive environmental change rate index is obtained by summing various change rates, then the formula for calculating the forgetting factor is, for example: in, Forgetting factor; It is the first forgetting factor; It is the second forgetting factor; It is the sensitivity coefficient. The larger the value, the faster the forgetting factor decays with the rate of environmental change. The value range of is (0, ∞), and the specific value can be determined based on historical data. Satisfy when When =0.5N, Attenuation to ( Nearly 2 / 2 (half-life method), where N is the amount of environmental data. It is a non-linear mapping term, which can ensure that the rate of environmental change changes smoothly over time, and conforms to the logical idea that the more drastic the change, the faster it should be forgotten. It is a standardized range, ensuring that the values ​​converge to [ ]. ]middle.

[0040] In another alternative embodiment, if the comprehensive environmental change rate index is obtained by weighting the various change rates, then the formula for calculating the forgetting factor is, for example: The meanings of each letter are explained above and will not be repeated here.

[0041] Specifically, when the comprehensive environmental change rate index approaches its minimum value of 0, the forgetting factor approaches its maximum forgetting factor of 0.99; when the comprehensive environmental change rate index approaches its maximum value of 1, the forgetting factor approaches its minimum forgetting factor of 0.8. The magnitude of the sensitivity coefficient determines the rate at which the forgetting factor decays from its maximum value to its minimum value. The larger the sensitivity coefficient, the more sensitive the forgetting factor is to changes in the comprehensive environmental change rate index, and the faster it decays.

[0042] Step S204: Based on the forgetting factor, perform an exponentially weighted moving average of the prediction error between historical prediction data and the current time to obtain the current window score of each candidate data window.

[0043] Historical prediction data refers to the score of each candidate data window at the previous time step, reflecting the overall prediction performance of that candidate data window up to the previous time step. Exponentially weighted moving average is a recursive weighted averaging method characterized by an exponential decay in the weight of historical data over time, meaning that recent data has a greater impact on the current result than older data. This application, through exponentially weighted moving average, effectively utilizes historical prediction performance information and dynamically adjusts the retention level of historical information based on the forgetting factor, thereby achieving a smooth and sensitive characterization of the score's prediction accuracy for the current window.

[0044] For example, this application may calculate the current window score of the candidate data window using the following formula: in: It is the current window score of window i at time t; It is the forgetting factor at time t; It is the score of window i at the previous time step; It is the prediction error obtained by window i at time t.

[0045] For example, this application can be used to update the current score of a candidate window with a length of 1 hour. If the score at the previous moment is 0.5, the prediction error at the current moment is 0.2, and the forgetting factor calculated based on environmental changes is 0.9, then the current score of the window is updated to 0.9×0.5+(1-0.9)×0.2=0.47. Through this recursive calculation, the scores of all candidate windows can be continuously refreshed.

[0046] In an optional embodiment, such as Figure 3 As shown, Figure 3 An optional embodiment of the steps provided in this application for correcting the current window score includes: Step S301: Detect whether there are transient event features in the multi-source environmental data. Transient event features include at least one environmental factor whose rate of change exceeds an abnormal threshold. Transient event characteristics can be used to characterize non-stationary and drastic changes in the crop growth environment within a short period of time. The detection logic can, for example, involve receiving real-time multi-source environmental data, analyzing the rate of change of each environmental factor, and identifying the presence of abnormal mutation signals. Specifically, this can be implemented by calculating the rate of change of each environmental factor at the current moment and comparing it with a preset anomaly threshold. If the rate of change of at least one environmental factor exceeds the anomaly threshold, a transient event characteristic is determined to exist.

[0047] Taking carbon dioxide concentration as an example, under normal greenhouse conditions, the change in carbon dioxide concentration per minute typically does not exceed 50 ppm. However, when a door or window is opened, the carbon dioxide concentration may drop sharply by hundreds of ppm within minutes, with an environmental factor change rate exceeding 100 ppm per minute. The computer equipment can compare the environmental factor change rate corresponding to the carbon dioxide concentration with its abnormal threshold, and simultaneously compare the comprehensive environmental change rate index with a preset comprehensive change rate abnormal threshold. When the environmental factor change rate corresponding to the carbon dioxide concentration exceeds its abnormal threshold and the comprehensive environmental change rate exceeds the comprehensive change rate abnormal threshold, a transient event is determined to exist, and the start time of the event is recorded. To ensure the reliability of the detection, the system can also be set to trigger event determination only after multiple consecutive sampling times meet the conditions, to eliminate misjudgments caused by instantaneous sensor noise or single-point fluctuations; this application does not limit this.

[0048] Step S302: When transient event characteristics are detected, an event impact coefficient is generated based on the transient event characteristics, and the corresponding sensitivity weight is determined based on the window length of each candidate data window. The event impact coefficient is used to quantify the disturbance intensity of a transient event on the window scoring system. It receives information on the rate of change of environmental factors contained in the transient event characteristics and outputs a value of impact intensity that decays over time. A specific implementation involves constructing a time-exponential decay function, which is activated at the moment the event occurs and determines the initial impact intensity based on the severity of the current environmental change (i.e., the magnitude of the rate of change). As time progresses, the event impact coefficient gradually decreases according to a preset decay rate until it reaches zero, thus simulating the gradual process of the transient event's impact.

[0049] In one optional implementation, the event impact coefficient is calculated by combining the current overall environmental change rate with a time decay factor. In the early stages of an event, the environmental factor change rate is high, resulting in a high event impact coefficient, indicating that the current environment is in a period of severe turbulence, and the correction force for the window scoring is significant. As time progresses, if no further drastic changes occur in the environment, the event impact coefficient decays exponentially, indicating that the event's impact gradually weakens and the window scoring mechanism gradually returns to normal. Through this dynamically generated event impact coefficient, the system can accurately match the lifecycle of transient events, achieving precise disturbance measurement.

[0050] Sensitivity weight can be considered as an adjustment coefficient for the correction magnitude determined by the window length of each candidate data window. Longer candidate data windows contain more historical data and are more susceptible to transient events, thus justifying a larger sensitivity weight. Sensitivity weight plays a crucial role in quantifying the impact of events within the entire correction mechanism. It transforms qualitative event judgments into calculable correction amounts and applies differentiated corrections based on the vulnerability of different windows. When a transient event occurs, the prediction error of all windows increases sharply simultaneously, and longer windows are more affected by outdated data from before the event. Applying the same correction magnitude to all windows fails to accurately reflect the differences in the degree of impact of the event on different windows.

[0051] In practical implementation, after detecting transient event characteristics, the event impact coefficient can be calculated first. The event impact coefficient is determined by the current comprehensive environmental change rate and the time decay factor. Specifically, the computer equipment uses the current comprehensive environmental change rate as the base intensity value of the event impact, which reaches its peak at the moment the event occurs. Subsequently, the system introduces an exponential decay function, causing the impact coefficient to gradually decrease over time. The decay rate is controlled by a preset time decay constant, which can be set according to the typical duration of the transient event. For example, the impact of a door or window opening event typically lasts five to fifteen minutes, so the time decay constant can be set to ten minutes. For each sampling moment after the event occurs, the time difference from the event's start time to the current moment is calculated. This time difference is divided by the time decay constant, and the negative exponential value is multiplied by the base intensity value to obtain the event impact coefficient at the current moment.

[0052] Simultaneously, the computer device also calculates a sensitivity weight for each candidate data window: obtaining the length value of each candidate data window and determining the maximum window length among all candidate data windows. For each candidate data window, the sensitivity weight can be obtained by dividing the window length of the candidate data window by the maximum window length. The value of this weight ranges from 0 to 1; the longer the window length, the closer the weight is to 1, and the shorter the window length, the closer the weight is to 0.

[0053] Step S303: The current window score of each candidate data window is corrected using the event impact coefficient and sensitivity weight. The longer the candidate data window is, the greater the correction magnitude of the current window score.

[0054] The scoring correction refers to applying additional adjustments to the current window score generated in the above steps to reflect the differentiated impact of transient events on the prediction accuracy of different candidate data windows. This application, by actively intervening in the current window score, can avoid window selection oscillations or switching delays caused by transient events. In conventional dynamic scoring mechanisms, changes in the current window score depend on the gradual accumulation of prediction errors; for sudden disturbances like transient events, the adjustment of the current window score has a certain lag. Through this direct correction step, the computer device can penalize the affected window score early in the event's occurrence, especially imposing a greater penalty on long windows, thereby prompting the computer device to quickly switch to a shorter window that is more suitable for the environment during the event.

[0055] For example, this application can correct the current window score based on the following formula: in: This is the rating for the current window; It is the rate of change of carbon dioxide at the current moment; It is the abnormal threshold of the rate of change of carbon dioxide; It is the abnormal threshold of the comprehensive environmental change rate; It is a comprehensive environmental change rate indicator; It refers to the current moment; It is the first moment when the activation function value is 1. It is the time decay constant; This is the current window length; This is the maximum candidate window length. When... When less than 0, take 0; when If the value is greater than 0, take the value 1.

[0056] It is an activation item. Based on the above description, it can be known that when When the value is greater than 0, the value is 1; when the value is less than 0, the value is 0. Since the most significant characteristic of window / door opening events is a sudden increase in carbon dioxide, accompanied by a period of sustained environmental turbulence, the decision to activate correction is made by simultaneously judging whether the rates of change in carbon dioxide and the overall environment exceed preset anomaly thresholds. The carbon dioxide and overall environmental thresholds can be obtained by averaging the rate of change data collected from multiple actual window / door opening experiments or by taking the 99th percentile of the environmental and carbon dioxide change rates from historical data.

[0057] This is the impact factor, characterizing the intensity of the impact. After an event occurs, the environment gradually returns to stability over time; this process follows an exponential decay pattern, hence the use of... The function, and as t increases, Gradually decreasing. Among them... Its function is to control the rate attenuation over time, and the duration of the effect is generally between 5 and 15 minutes. Therefore, the maximum value of this scheme is 15.

[0058] For example, this application can involve a computer device monitoring environmental data in real time during operation. When it detects that the rate of change in carbon dioxide concentration drops sharply beyond a preset threshold (e.g., 10%) within one minute, it determines that a transient event of opening doors and windows has occurred. At this time, the computer device calculates the initial event impact coefficient based on the current comprehensive environmental change rate and obtains the sensitivity weight of each candidate window (e.g., a 30-minute window, a 1-hour window, a 2-hour window, etc.). For a long window of 2 hours, due to its longer window length, the sensitivity weight is set higher, and the correction amount calculated by the system is also larger; for a short window of 30 minutes, the sensitivity weight is lower, and the correction amount is smaller. The computer device adds these correction amounts to the corresponding original window scores, causing the scores of long windows to rise significantly in a short period of time (i.e., performance deteriorates), so that in the subsequent target window selection process, the system can quickly switch to a short window to more accurately predict the crop transpiration demand under the changed environment.

[0059] This application achieves dynamic correction of the current window score for candidate data windows by detecting transient event characteristics, generating event impact coefficients, and determining sensitivity weights. Leveraging the time-dependent decay of the event impact coefficient, the dynamic process of a transient event from occurrence to end is simulated. By using sensitivity weights determined based on window length, differentiated processing is applied to address the varying degrees of dependence on historical information by different windows, thus reasonably suppressing long windows during transient events. This solves the problems of window selection oscillation and switching delays caused by the lag in historical data from long windows when facing sudden environmental changes such as opening greenhouse doors and windows in existing technologies. It effectively improves the system's robustness and response speed in complex dynamic environments, ensuring the accuracy of transpiration prediction and the timeliness of environmental control.

[0060] Step S50: Select a target window from multiple candidate data windows based on the current window score, and perform environmental regulation based on the evapotranspiration prediction value corresponding to the target window.

[0061] The target window can refer to the candidate data window that is selected from all candidate data windows and has the best score in the current window (e.g., the lowest or highest score, depending on the score definition).

[0062] Environmental regulation can refer to the operation of sending control instructions to the implementing agency based on the forecast results to adjust the crop growth environment (such as turning on irrigation, adjusting ventilation equipment, etc.). This step compares the current window scores of all candidate windows, selects the target window according to the preset optimization strategy, and uses the transpiration forecast value corresponding to the window as the final decision basis to output the regulation instruction.

[0063] Specifically, this application can be implemented in a greenhouse control system by comparing the scores of windows of different lengths to select the window that best adapts to the rhythm of current environmental changes. The predicted value of this window is then compared with a dynamic control threshold. When the predicted value exceeds the threshold, irrigation or ventilation equipment is automatically triggered.

[0064] This application dynamically adjusts the weighting of historical forecast data in the window scoring by introducing the rate of change of environmental factors. This allows the scoring mechanism to adaptively match the pace of environmental change: when environmental changes are drastic, reducing the weight of historical data quickly eliminates the influence of outdated information, thereby improving the window selection's response speed to sudden environmental changes; when the environment is stable, increasing the weight of historical data effectively utilizes the statistical regularities of long-term data to smooth random noise, thus ensuring the stability of window selection. The target window selected based on this scoring consistently provides the most accurate evapotranspiration forecast, thereby achieving precise environmental control and solving the technical problem in existing technologies where fixed windows cannot simultaneously balance response speed and anti-interference capabilities.

[0065] In an optional embodiment, such as Figure 4As shown, Figure 4 An optional method embodiment of the intelligent monitoring method for crop cultivation and growth environment provided in an exemplary embodiment of this application includes the following steps: Step S401: Compare the predicted evaporation value corresponding to the target window with the dynamic control threshold; The dynamic control threshold refers to a critical value used to trigger the action of the equipment, which is dynamically adjusted based on multiple factors such as crop growth stage, soil moisture conditions, and weather forecast information. This threshold is not a fixed constant but can be a variable that can adaptively adjust with changes in crop water requirements and environmental conditions.

[0066] The comparison between the predicted transpiration value and the dynamic control threshold refers to comparing the predicted transpiration value output by the target window selected in the above steps with the current dynamic control threshold to determine whether the preset activation conditions for the execution equipment have been met or exceeded. This application, by setting a dynamic control threshold, can flexibly adjust the control sensitivity according to the actual water requirements of the crop, avoiding over-irrigation or under-irrigation caused by a fixed threshold. When the predicted transpiration value is lower than the threshold, it indicates that the current crop transpiration demand is still within a controllable range, and there is no need to activate the control equipment; when the predicted transpiration value reaches or exceeds the threshold, it indicates that the crop's water demand has approached or exceeded the critical point, and the control equipment needs to be activated promptly to meet the crop's growth needs.

[0067] For example, the computer device may maintain a dynamic control threshold calculation module. At the beginning of each control cycle, this module reads the current crop growth stage information, such as seedling stage, flowering stage, and fruiting stage, as the water requirements of crops at different growth stages differ significantly. Simultaneously, the module reads data from multi-layer soil moisture sensors deployed in the soil to obtain the current volumetric water content of the root zone soil. Based on this, the module also connects to a weather forecast interface to obtain predicted data such as light intensity, temperature, and wind speed for the next few hours. The dynamic control threshold calculation module weights and adjusts the base threshold according to a crop coefficient, soil moisture correction coefficient, and weather correction coefficient preset for the crop type, generating the dynamic control threshold for the current moment. This dynamic control threshold is expressed in units of transpiration rate, such as millimeters per hour or grams per square meter per minute. Subsequently, the control decision module obtains the predicted transpiration value corresponding to the target window from the window selection module, compares this predicted value with the dynamic control threshold, and records the comparison result for use in subsequent steps.

[0068] Step S402: When the predicted evaporation value exceeds the dynamic control threshold, the first execution device is triggered to perform control operations and generate control commands. The first executing device can refer to a basic actuator deployed on-site in the greenhouse that can directly change environmental parameters, such as a solenoid valve for irrigation, a window opener for ventilation, or a shade net drive motor for adjusting light. The logic of this step is to receive the judgment result output from the comparison step. Once it is confirmed that the transpiration demand exceeds the threshold, a control signal is immediately sent to the first executing device to initiate the corresponding physical operation. Simultaneously, based on the control objectives and strategies, control instructions are generated to coordinate other equipment or record control information. For example, when an excessively high transpiration prediction is detected, the system automatically triggers the opening of water pumps or valves in the irrigation system to replenish water; or, when the prediction indicates that the environment is overheated, causing abnormal transpiration, the ventilation fan is triggered to start. At the same time, the system generates control instructions containing information such as the control type, intensity, and start time.

[0069] This application ensures the immediacy of on-site physical operations by generating control commands simultaneously with triggering the first execution device, and provides a standardized command basis for subsequent multi-device collaboration or status synchronization, thereby realizing the rapid transformation from data decision-making to physical action.

[0070] Step S403: Send the control command to the second execution device to perform the control operation; The second executing device can refer to other actuators that need to perform coordinated actions, supplementary operations, or respond to complex control logic after the first executing device performs basic operations. Examples include a fertilizer applicator working with a drip irrigation system, a wet curtain fan system working with a ventilation fan, or multiple distributed control units used for regional environmental linkage. The function of this step is to send the control commands generated in the previous step to the designated second executing device via network transmission or bus communication, enabling it to parse the command content and execute the corresponding environmental parameter adjustment operations.

[0071] For example, the computer equipment sends the generated control instructions to the circulating fan and fertilizer pump in the greenhouse. After receiving the instructions, the circulating fan starts to rotate to evenly distribute humidity and temperature, and the fertilizer pump releases nutrient solution according to the ratio parameters in the instructions, thereby achieving precise control of water and fertilizer integration based on the water replenishment of the first execution device.

[0072] This application, by sending control commands to a second execution device and utilizing the collaborative work of multiple devices, compensates for the shortcomings of the control capabilities of a single device, achieves multi-dimensional and three-dimensional precise control of the crop growth environment, and enhances the overall effectiveness of the environmental control system.

[0073] Step S404: Receive environmental feedback data sent by the second execution device based on the control command, and adaptively calibrate the dynamic control threshold based on the environmental feedback data.

[0074] The environmental feedback data refers to data collected by the second actuator's built-in sensors or environmental monitoring network after performing the control operation, reflecting the actual changes in environmental parameters. Examples include changes in soil moisture after irrigation and decreases in air temperature and humidity after ventilation. The processing logic of this step involves receiving the execution result data from the second actuator, analyzing the actual environmental effects of the control operation, and adjusting the system's internal dynamic control thresholds based on the deviation between the actual effects and the expected targets to better suit the actual needs of the crop under the current environment.

[0075] For example, the computer device receives post-execution environmental feedback data sent by the second execution device, such as monitoring that the air humidity increased by 5% after irrigation. Based on this, it calculates the actual inhibitory effect of the environmental change on the transpiration rate, and then fine-tunes the dynamic control threshold through gradient descent or proportional integral algorithms, and accordingly lowers or raises the threshold for triggering the control next time to adapt to the environmental response characteristics under the current season or weather conditions.

[0076] This application introduces a threshold adaptive calibration mechanism based on environmental feedback data, and uses the actual effect data after regulation to back-optimize the decision benchmark. This solves the problem that fixed thresholds cannot adapt to long-term environmental drift and equipment aging, thereby enabling the intelligent monitoring system to have the ability to learn and evolve on its own, ensuring that it maintains optimal regulation performance in different growth cycles and external environments.

[0077] It should be noted that although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all of the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart may be performed in a different order.

[0078] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for intelligent monitoring of crop cultivation and growth environment, characterized in that, The method includes: Receive multi-source environmental data on crop growth environment, and divide the multi-source environmental data into multiple candidate data windows with different time scales; Each of the candidate data windows is input into the target transpiration prediction model to obtain the transpiration prediction value corresponding to each of the candidate data windows; The current window score of each candidate data window is determined based on the historical prediction data of each candidate data window and the rate of change of environmental factors. The rate of change of environmental factors is used to dynamically adjust the weight ratio of the historical prediction data in the current window score. Based on the current window score, a target window is selected from multiple candidate data windows, and environmental regulation is carried out based on the evapotranspiration prediction value corresponding to the target window.

2. The method according to claim 1, characterized in that, The method further includes preprocessing the multi-source environmental data, which includes outlier removal, missing value completion, and normalization of the multi-source environmental data.

3. The method according to claim 1, characterized in that, The process of determining the current window score for each candidate data window based on historical prediction data and the rate of change of environmental factors includes: Based on the historical prediction data of each candidate data window, the prediction error of each candidate data window is calculated; Based on the multi-source environmental data, the change rates of multiple environmental factors are determined, and a comprehensive environmental change rate index is constructed based on the change rates of each environmental factor. The forgetting factor is dynamically adjusted based on the comprehensive environmental change rate index. Based on the forgetting factor, an exponentially weighted moving average is applied to the prediction error between the historical prediction data and the current moment to obtain the current window score for each candidate data window.

4. The method according to claim 3, characterized in that, The dynamic adjustment of the forgetting factor based on the comprehensive environmental change rate index includes: The comprehensive environmental change rate index is input into a nonlinear mapping function, and the forgetting factor is output. The nonlinear mapping function is configured such that the forgetting factor decreases monotonically as the comprehensive environmental change rate index increases, and the range of the forgetting factor is limited to between a first forgetting factor and a second forgetting factor. The first forgetting factor is the lower limit value of the forgetting factor set when the crop growth environment is in a stage of drastic change, and the second forgetting factor is the upper limit value of the forgetting factor set when the crop growth environment is in a stable stage.

5. The method according to claim 1, characterized in that, The method further includes: The system detects whether transient event characteristics exist in the multi-source environmental data. The transient event characteristics include at least one environmental factor whose rate of change exceeds an anomaly threshold. When the transient event characteristics are detected, an event impact coefficient is generated based on the transient event characteristics, and the corresponding sensitivity weight is determined based on the window length of each candidate data window. The current window score of each candidate data window is corrected using the event impact coefficient and the sensitivity weight, wherein the larger the window length of the candidate data window, the greater the correction magnitude of the current window score.

6. The method according to claim 1, characterized in that, The method further includes: Compare the predicted transpiration value corresponding to the target window with the dynamic control threshold; When the predicted transpiration value exceeds the dynamic control threshold, the first execution device is triggered to perform a control operation and generate a control command. The control command is sent to the second execution device to perform the control operation; Based on the control command, the system receives environmental feedback data sent by the second execution device and adaptively calibrates the dynamic control threshold based on the environmental feedback data.

Images