Crop irrigation short-term prediction method and system based on fuzzy control

Abstract

The invention discloses a crop irrigation short-term prediction method and system based on fuzzy control, and relates to the technical field of intelligent agriculture and irrigation control, and the method comprises the steps: collecting weather, multi-depth soil moisture content and field time sequence data of a crop growth stage, and obtaining irrigation feedback; processing the data, and inputting a time sequence prediction model to output water demand prediction values of one to three days and confidence obtained by mapping historical verification errors; fuzzifying the water demand predicted value, the real-time root zone moisture content and the growth stage index, performing weighted deduction and fuzzification based on a fuzzy rule base with a rule dynamic weight, and outputting planned irrigation duration or irrigation amount; and calculating the actual water consumption and updating the dynamic weight of the rule in the post-irrigation evaluation time window according to the water balance and the root zone soil moisture reserve variation. Through confidence risk suppression and post-irrigation closed-loop self-tuning, the stability of short-term prediction driving irrigation decision is improved, the over-irrigation and over-seepage risks are reduced, and the utilization efficiency of irrigation water is improved.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture and irrigation control technology, specifically to a short-term crop irrigation prediction method and system based on fuzzy control. Background Technology

[0002] The core objective of crop irrigation management is to improve irrigation water use efficiency and reduce secondary problems caused by deep seepage, runoff loss, and over-irrigation, while meeting crop water requirements. With the widespread adoption of field sensors and automated irrigation equipment, intelligent irrigation based on meteorological, soil moisture, and crop growth information is gradually replacing experience-based irrigation. However, in practical applications, existing technologies still have the following shortcomings, making it difficult to achieve stable and reliable short-term irrigation plan outputs in complex environments.

[0003] First, there is insufficient coupling between short-term water demand forecasting and irrigation decisions. Existing schemes often estimate water demand using a single indicator or empirical model and directly map the estimation results to irrigation duration or irrigation volume, lacking unified modeling of multi-source time-series data and the ability to predict short-term fluctuations over the next one to three days. Especially during periods of rapid weather changes, uncertain precipitation, or transitions in growth stages, water demand changes exhibit significant nonlinear and time-lag characteristics. Single estimations can easily lead to the accumulation of deviations in irrigation plans, resulting in under-irrigation or over-irrigation.

[0004] Secondly, the uncertainty in predictions lacks quantification and control mechanisms. Even when existing methods incorporate data-driven prediction models, they often only output point predictions, failing to provide a reliable representation for control decisions. When training data for the prediction model is insufficient or the scenario shifts, prediction errors may increase, but irrigation decisions are still executed at the same intensity, amplifying uncertainty into control risks. The lack of mechanisms to explicitly incorporate prediction reliability into the decision-making process and suppress decision intensity is one of the main reasons why smart irrigation is "prone to instability and difficult to promote."

[0005] Third, the irrigation closed-loop feedback correction is insufficient, making it difficult for rules to adapt to long-term field changes. Soil texture, root distribution, irrigation methods, and seasonal climate constantly change, resulting in different root zone watering effects at different stages with the same irrigation volume. Existing technologies, such as fuzzy control or rule-based control, often use fixed rule bases and fixed membership functions, lacking a systematic evaluation and update mechanism based on post-irrigation response, making it difficult to maintain rule effectiveness over long-term operation. Furthermore, they lack characterization of the process risks of root zone retention and deep penetration, making it difficult to promptly identify hidden problems such as "insufficient root zone watering but increased deep leakage." Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a short-term crop irrigation prediction method and system based on fuzzy control to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a short-term crop irrigation prediction method based on fuzzy control, comprising:

[0008] Collect multi-dimensional time-series data in the field, including at least meteorological data, soil moisture content at multiple depths and crop growth stage data, and obtain feedback on irrigation flow or irrigation volume.

[0009] Perform time alignment, outlier removal, missing data imputation, and normalization on time-series data;

[0010] The processed time series data is input into a time series prediction model with a preset historical time window, and the predicted crop water requirement and confidence level for one to three days are output. The confidence level is obtained by monotonic mapping and normalization based on the historical verification error of the time series prediction model.

[0011] The predicted water demand, real-time root zone soil moisture content, and growth stage index are fuzzified. Weighted reasoning and defuzzification are performed based on a fuzzy rule base with dynamic weights and rules. The output is the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment.

[0012] Within a preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in soil moisture storage in the root zone from the sum of irrigation volume and precipitation. The dynamic weights of the rules in the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, which is used to adaptively correct the planned irrigation duration or irrigation volume for subsequent irrigation.

[0013] The present invention is further configured such that the meteorological data includes temperature, humidity, wind speed, radiation and precipitation; the multi-depth soil moisture content includes at least the volumetric moisture content of the shallow, middle and deep soil layers in the root zone; the crop growth stage data includes growth stage identifiers or growth stage indices; and the irrigation flow or irrigation volume feedback includes at least one of the following: irrigation start time, duration, valve opening / closing status, and valve opening degree.

[0014] The present invention is further configured such that anomaly removal includes joint detection of mutation screening based on threshold of change magnitude between adjacent time periods and outlier discrimination based on historical distribution; missing data filling includes completion based on time-series interpolation or near-neighbor interpolation based on similar days; and normalization is based on the dimensional unification of sliding time window statistics, so that meteorological data, soil moisture content and growth stage index can be used as inputs to the time-series prediction model.

[0015] The present invention is further configured such that the real-time root zone soil moisture content is obtained by mapping the soil moisture content at multiple depths through the root zone depth. The root zone depth mapping includes: determining the upper boundary depth and lower boundary depth of the root zone; and fusing the moisture content at each depth falling within the root zone range according to a preset depth attenuation function to obtain the real-time root zone soil moisture content.

[0016] The present invention is further configured such that the time series prediction model adopts a bidirectional time series network with an attention mechanism, the input is a meteorological data sequence, a multi-depth soil moisture content sequence and a growth stage index sequence within a preset historical time window, and the output is the crop water requirement prediction value corresponding to one to three days respectively; the confidence level is obtained by monotonically mapping and normalizing the historical verification error obtained by the time series prediction model in the training and verification stage, and is set as the risk suppression factor for subsequent fuzzy inference.

[0017] The present invention is further configured such that the input variables of the fuzzy rule base include at least the predicted water demand value, the real-time root zone soil moisture content and the growth stage index, and each is provided with a fuzzy language set of no less than three levels; the input variables are fuzzified using triangular membership functions or Gaussian membership functions; the output variable is the planned irrigation duration or irrigation volume, and the continuous output value is obtained by defuzzifying using the centroid method.

[0018] The present invention is further configured such that the fuzzy rule base includes a set of rules covering multiple water demand levels, multiple soil moisture states and multiple growth stage combinations, and the number of rules is not less than twenty-seven; each rule is associated with a rule dynamic weight, and the rule dynamic weight, together with the corresponding rule trigger strength, determines the contribution of the rule during weighted inference; when the rule dynamic weight is lower than a preset lower limit, the rule dynamic weight is frozen or the rule is temporarily disabled.

[0019] The present invention is further configured to construct the root zone water retention potential within a preset evaluation time window after each irrigation. The determination of the root zone water retention potential includes: determining the root zone depth range and selecting soil moisture content at multiple depths within the root zone; calculating the increase in root zone soil moisture storage from the end of irrigation to the end of the evaluation time window, and characterizing the root zone retention ratio by the ratio of the increase in root zone soil moisture storage to the sum of the irrigation amount and precipitation; obtaining the root zone water retention potential by monotonic mapping and normalization of the retention ratio; and using the root zone water retention potential as an additional input for fuzzy inference or as a correction factor for the planned irrigation duration or irrigation amount, wherein when the root zone water retention potential is lower than a preset threshold, the planned irrigation duration is reduced or the planned irrigation duration is divided into multiple segments for execution.

[0020] The present invention is further configured to construct a penetration order deviation within a preset evaluation time window after each irrigation. The determination of the penetration order deviation includes: determining the response time when the soil moisture content at each depth reaches a preset increase threshold from the end of irrigation for soil moisture content at multiple depths; establishing a reference penetration order with increasing response times from shallow to deep; comparing the measured response time sequence with the reference penetration order, generating a deviation degree based on the number of reverse occurrences and the degree of abrupt change in the response time difference between adjacent depths, and normalizing it to obtain the penetration order deviation; wherein, when a certain depth does not reach the preset increase threshold within the preset evaluation time window, the end time of the preset evaluation time window is taken as the response time of that depth, or the depth is marked as non-responsive and included in the deviation degree according to a preset penalty term; the penetration order deviation is used as an over-infiltration risk constraint in fuzzy reasoning and rule dynamic weight update, wherein when the penetration order deviation exceeds the preset threshold, the rule dynamic weight corresponding to high irrigation volume or long irrigation duration is reduced and a protection strategy of reducing valve opening or shortening single irrigation duration is triggered.

[0021] This invention also provides a short-term crop irrigation forecasting system based on fuzzy control, used to implement the above-mentioned short-term crop irrigation forecasting method based on fuzzy control, comprising: a processor, a memory, and a communication interface, wherein the memory stores a program executable by the processor, and the processor executes the program to implement the following functional units:

[0022] Multi-source acquisition module: Collects multi-dimensional time-series data in the field, including at least meteorological data, soil moisture content at multiple depths and crop growth stage data, and obtains feedback on irrigation flow or irrigation volume;

[0023] Preprocessing module: performs time alignment, anomaly removal, missing data imputation, and normalization on time-series data;

[0024] Short-term forecasting module: Input the processed time series data into the time series forecasting model with a preset historical time window, and output the crop water requirement forecast value and confidence level for one to three days. The confidence level is obtained by monotonic mapping and normalization based on the historical verification error of the time series forecasting model.

[0025] Fuzzy decision module: It fuzzifies the water demand forecast, real-time root zone soil moisture content and growth stage index, performs weighted reasoning based on a fuzzy rule base with dynamic weights and defuzzifies it, and outputs the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment.

[0026] Feedback self-adjustment module: Within the preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in soil moisture storage in the root zone from the sum of irrigation volume and precipitation. The dynamic weights of the rules in the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, which is used to adaptively correct the planned irrigation duration or irrigation volume in the future.

[0027] This invention provides a method and system for short-term crop irrigation forecasting based on fuzzy control. The method collects multi-dimensional time-series data from the field, including at least meteorological data, soil moisture content at multiple depths, and crop growth stage data, to obtain irrigation flow or water volume feedback. The time-series data undergoes time alignment, anomaly removal, missing data imputation, and normalization. The processed time-series data is input into a time-series forecasting model with a preset historical time window, outputting crop water requirement forecasts and confidence levels for one to three days. The confidence level is obtained by monotonic mapping and normalization of the historical validation error of the time-series forecasting model. The water requirement... The predicted values, real-time root zone soil moisture content, and growth stage index are fuzzified. Weighted inference is then performed based on a fuzzy rule base with dynamic rule weights, followed by defuzzification. The output is the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment. Within a preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in root zone soil moisture storage from the sum of irrigation volume and precipitation. The dynamic rule weights of the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, used for adaptive correction of subsequent planned irrigation duration or irrigation volume. The beneficial effects include:

[0028] 1. By uniformly preprocessing meteorological data, soil moisture content at multiple depths, and crop growth stage data, and inputting time series data with preset historical time windows into the time series prediction model, the predicted crop water demand values ​​for the next one to three days are obtained. This enables irrigation plans to no longer rely on single-point experience estimations, but to reflect the impact of short-term meteorological fluctuations and soil moisture status on water demand changes, thereby improving the foresight and stability of irrigation plans.

[0029] 2. In fuzzy inference, dynamic weights are assigned to each rule, and the actual water consumption is calculated within a preset evaluation time window after irrigation. The dynamic weights are updated based on the deviation between the actual water consumption and the predicted water demand, thus forming a closed loop of "prediction-decision-execution-evaluation-correction". Compared with schemes using fixed rules or fixed membership functions, this invention can continuously correct the rule contribution, reduce long-term drift caused by soil conditions, root distribution, and seasonal changes, and improve the consistency of irrigation plan output.

[0030] 3. Within the assessment time window, the actual water consumption is obtained by subtracting the change in root zone soil moisture storage from the sum of irrigation and precipitation. This ensures that the assessment results are consistent with the changes in root zone moisture, avoids deviations caused by directly using changes in water content for water calculation, improves the interpretability and feasibility of feedback updates, and provides a more reliable correction basis for the dynamic weight update of rules.

[0031] 4. Construct a root zone water retention potential to measure the effective retention level of water in the root zone, and construct a penetration order deviation to measure abnormal water response order and the risk of over-infiltration. When the root zone water retention potential is low or the penetration order deviation exceeds the threshold, the planned irrigation intensity can be reduced or protective actions such as reducing valve opening and shortening the duration of a single irrigation can be triggered. This reduces the risk of deep seepage and over-irrigation while meeting the crop's water requirements, improving safety and scalability.

[0032] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0034] Figure 1 A flowchart illustrating a short-term crop irrigation forecasting method based on fuzzy control is shown as an exemplary embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram illustrating the structure of a short-term crop irrigation prediction system based on fuzzy control, as an exemplary embodiment of the present invention. Detailed Implementation

[0036] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0037] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0038] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0039] Example 1:

[0040] A short-term crop irrigation forecasting method based on fuzzy control, such as Figure 1 As shown, it includes:

[0041] Collect multi-dimensional time-series data in the field, including at least meteorological data, soil moisture content at multiple depths and crop growth stage data, and obtain feedback on irrigation flow or irrigation volume.

[0042] Perform time alignment, outlier removal, missing data imputation, and normalization on time-series data;

[0043] The processed time series data is input into a time series prediction model with a preset historical time window, and the predicted crop water requirement and confidence level for one to three days are output. The confidence level is obtained by monotonic mapping and normalization based on the historical verification error of the time series prediction model.

[0044] The predicted water demand, real-time root zone soil moisture content, and growth stage index are fuzzified. Weighted reasoning and defuzzification are performed based on a fuzzy rule base with dynamic weights and rules. The output is the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment.

[0045] Within a preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in soil moisture storage in the root zone from the sum of irrigation volume and precipitation. The dynamic weights of the rules in the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, which is used to adaptively correct the planned irrigation duration or irrigation volume for subsequent irrigation.

[0046] Specifically, several irrigation control units are divided within the crop planting area, with each control unit corresponding to at least one irrigation branch and a set of sensor data collection points. Multidimensional time-series data is collected within each irrigation control unit, including at least meteorological data, soil moisture content data at multiple depths, and crop growth stage data. Simultaneously, irrigation flow or irrigation volume feedback data corresponding to that control unit is acquired, and all data types are written into the same time-series data table according to timestamps for subsequent modeling and decision-making.

[0047] Furthermore, meteorological data is collected periodically by small field weather stations, with collection cycles set to minutes or hours. The collected meteorological data includes temperature, humidity, wind speed, radiation, and precipitation. Temperature and humidity characterize evapotranspiration-driven thermal and humid conditions, wind speed characterizes boundary layer transport conditions, radiation characterizes net radiation input intensity, and precipitation characterizes natural water replenishment input. The weather stations upload the data to the controller or edge computing terminal via wired or wireless means, forming a continuous meteorological time series.

[0048] Multi-depth soil moisture content is collected by soil moisture sensors deployed at different depths within the root zone, including at least the volumetric water content of the shallow, middle, and deep layers of the root zone. During implementation, three or more depth sampling points can be arranged on the same vertical profile. Shallow sensors are deployed at the upper depth near the active root water absorption zone, middle sensors at the middle depth of the root zone, and deep sensors at the depth near the lower boundary of the root zone to reflect irrigation infiltration and water redistribution processes. The volumetric water content at each depth is output with the same sampling period and uploaded with a timestamp, forming a multi-depth moisture content time series.

[0049] Crop growth stage data is used to describe the differences in water requirements of crops at different growth stages, and can be provided in two ways: growth stage identifiers or growth stage indices. In one implementation, growth stage identifiers are manually entered or synchronously written by an agricultural management platform, using discrete stage labels to represent the crop's growth stage. In another implementation, growth stage indices are calculated from historical accumulated temperature, sowing date, variety parameters, or canopy observation data, and are updated over time as continuous numerical values. Growth stage identifiers or growth stage indices are linked and stored with meteorological and soil moisture data, serving as important inputs for subsequent short-term water requirement forecasting and fuzzy inference.

[0050] Irrigation flow rate or irrigation volume feedback is collected from the irrigation execution end. In one embodiment, a flow meter is installed on the irrigation pipeline to collect instantaneous flow rate and integrate it to obtain the irrigation volume. Simultaneously, a valve controller records the irrigation start time and duration, and transmits the valve opening / closing status and valve opening degree. In another embodiment, if no flow meter is configured, at least one of the following is collected: irrigation start time, duration, valve opening / closing status, and valve opening degree. This data is then combined with the calibrated flow rate of the irrigation equipment or the valve opening degree-flow rate mapping relationship to calculate the irrigation flow rate or irrigation volume feedback. The above irrigation feedback data, together with multidimensional time-series data from the field, constitute a complete record of the same irrigation event, which is used for subsequent water demand prediction, irrigation plan generation, and post-irrigation evaluation and adaptive correction.

[0051] To ensure that meteorological data, soil moisture content at multiple depths, and crop growth stage data can be used as a unified input to drive subsequent time series prediction models, time alignment, anomaly removal, missing data imputation, and normalization are performed sequentially on the collected multi-source time series data, and the preprocessed sample sequences are output.

[0052] Because of differences in sampling periods and reporting delays among weather stations, soil sensors, and growth stage data sources, a unified time axis was first constructed, and each data source was mapped to this time axis. Specifically, the target time granularity was set to a preset sampling interval, and the raw data was aligned based on timestamps: for each target moment, the most recent observation value from each data source whose time difference with the target moment did not exceed a preset tolerance was selected as the corresponding value for that target moment; if no observation value was found within the tolerance range, the variable for that target moment was marked as missing. For growth stage indices or stage identifiers, a "preservation alignment" strategy was adopted, that is, the previous valid stage value was maintained within the time interval before the stage change occurred, to ensure that stage information was continuously available on the time axis.

[0053] Anomaly removal is performed on each aligned variable, consisting of two levels of joint detection:

[0054] First, mutation screening. The magnitude of change for each variable between adjacent target times is calculated and compared to a preset magnitude threshold. When the magnitude of change exceeds the threshold, the data point is marked as a candidate anomaly. The magnitude threshold can be set separately for different variables; for example, a stricter threshold for the rise / fall of soil moisture content can be set to capture non-physical jumps caused by sensor drift, transient interference, or communication jitter.

[0055] Second, outlier detection. For candidate anomalies, outlier detection is further performed based on historical distribution: In a historical sample set of the same growth stage or similar meteorological conditions, the historical distribution range of the variable is constructed (e.g., represented by quantile intervals or robust intervals). When a candidate anomaly falls outside the historical distribution range, it is confirmed as an outlier. When a candidate anomaly undergoes a sudden change but is still within the acceptable range of the historical distribution, the anomaly label is removed and the point is retained to avoid misjudging real physical changes such as sudden increases in rainfall or rapid infiltration after irrigation as anomalies.

[0056] For confirmed outliers, a removal process is performed, marking them as missing data and allowing them to proceed to the subsequent missing data filling process, thereby ensuring the continuity and reliability of the input sequence.

[0057] Hierarchical completion is performed on missing data, including completion based on time-series interpolation or completion based on nearest neighbor interpolation of similar days.

[0058] In one implementation, when the duration of a missing segment does not exceed a preset length, temporal interpolation is used to complete the segment, allowing for a smooth transition between adjacent valid points. Temporal interpolation can employ linear interpolation, piecewise hold interpolation, or gradually constrained interpolation, depending on the characteristics of the variable, to avoid introducing non-physical oscillations. For pulse-type variables such as precipitation, a "zero hold + event correction" strategy can be used to prevent the interpolation from fabricating precipitation events.

[0059] In another implementation, when the missing segment is prolonged, spans irrigation events, or exhibits a clear diurnal cycle, nearest neighbor interpolation based on similar days is used to complete the missing segment. Nearest neighbor interpolation based on similar days involves: selecting a set of candidate dates with consistent or similar growth stages from historical data; selecting nearest neighbor dates based on the similarity of meteorological conditions and the initial state of soil moisture content; using the variable sequence of the selected nearest neighbor dates in the corresponding time period as a reference, and aligning the amplitude with known observation points for the current day to fill the missing segment. Through near neighbor interpolation based on similar days, the reasonableness of the sequence morphology can be maintained even when the missing segment is prolonged or a strong periodic pattern exists.

[0060] After anomaly removal and missing data imputation, normalization is performed on each variable to ensure that meteorological data, soil moisture content, and growth stage indices of different dimensions can be used as input to the time-series prediction model at the same numerical scale. Normalization employs a dimensionless approach based on sliding time window statistics: for each variable, its statistical descriptive value is calculated within the sliding time window, and the variable value is mapped to a preset normalization interval based on this statistical descriptive value. The sliding time window is used to adapt to dimensional scale drift caused by seasonal and stage variations, thereby avoiding long-term shifts in the input distribution caused by using a fixed global scale.

[0061] In practical implementation, meteorological variables and growth stage indices can be mapped using robust central and robust discrete quantities within a sliding window, ensuring that extreme weather events do not significantly stretch the scale. Soil moisture content can be mapped using a sliding window incorporating its physical upper and lower limits, ensuring that the normalization result reflects short-term fluctuations without disrupting the relative relationship of moisture content. After normalization, a multi-source input sequence on a unified time axis is obtained, providing a stable input basis for the subsequent time-series forecast model to output water demand predictions and confidence levels for the next one to three days.

[0062] After completing time alignment, outlier removal, missing data imputation, and normalization of multi-source time series data, the processed time series data are used to construct samples according to a preset historical time window. Specifically, taking the current decision-making moment as a reference, the preset historical time window length is traced back, and meteorological data sequences, multi-depth soil moisture content sequences, and growth stage index sequences within this time window are extracted. These three types of sequences are then spliced ​​together in a unified time axis order to form historical feature sequence samples. When diurnal periodic differences exist, time position markers or stage position markers can be retained in the samples to enhance the model's ability to express periodicity and stages, thereby providing an input basis for subsequent short-term water demand forecasting.

[0063] The time-series forecasting model employs a bidirectional time-series network with an attention mechanism. This network simultaneously models the impact of previous moments on the current moment within a historical time window, as well as the supplementary constraints imposed by subsequent moments on key states. This allows the model to establish more stable correlations between changes in soil moisture content, meteorological drivers, and growth stages before and after irrigation events. The attention mechanism assigns different importance weights to different time segments within the historical time window, ensuring that key time segments more relevant to changes in water demand (such as the post-irrigation infiltration redistribution phase, periods of sudden rainfall increases, and phases of significant changes in radiation) contribute more to the model output, thereby enhancing the short-term forecast's responsiveness to key disturbances.

[0064] The model outputs predicted crop water requirements for the next one to three days. Specifically, the time-series prediction model extracts time-series representations from the input historical feature sequence and generates water requirement output sequences for multiple prediction steps, with each prediction step corresponding to a one-day water requirement prediction. The output water requirement predictions for the next one, two, and three days are used as the basis for generating short-term irrigation plans. Subsequently, these predictions, along with real-time root zone soil moisture content and growth stage indices, enter the fuzzy inference stage to obtain an executable planned irrigation duration or irrigation volume.

[0065] The confidence score is obtained by monotonically mapping and normalizing the historical validation error obtained during the training and validation phases of the time-series prediction model. Specifically, after model training is completed, the historical validation error of the model at each prediction step size is calculated using the validation dataset, and this error is used as a representation of the model's reliability. To ensure the consistency that "increased error leads to decreased confidence," a monotonically mapping relationship between error and confidence is constructed, so that smaller errors correspond to higher confidence and larger errors correspond to lower confidence. The mapped results are then normalized to obtain the confidence score value. The confidence score can be obtained separately for each prediction step size, or a unified confidence score can be obtained by uniformly processing the errors of multiple steps, both of which satisfy the purpose of serving as a risk constraint parameter for control decisions.

[0066] The confidence level is set as a risk suppression factor for subsequent fuzzy inference. Specifically, in the subsequent fuzzy inference stage, the confidence level participates in the inference process as an input parameter or as a modulation parameter for the dynamic weights of the rules: when the confidence level decreases, the contribution of the rules corresponding to high irrigation volume or long irrigation duration is reduced, or a suppressive correction is performed on the output planned irrigation duration / irrigation volume, so that the irrigation strategy converges in a conservative direction, thereby reducing the risk of prediction uncertainty being directly converted into excessive control; when the confidence level is high, the planned output is allowed to respond more fully to changes in water demand forecast, so as to improve the matching degree and execution efficiency of short-term irrigation plans.

[0067] After obtaining the water demand forecast for the next one to three days, a real-time root zone soil moisture content is further constructed. The water demand forecast, real-time root zone soil moisture content, and growth stage index are used as the core inputs for fuzzy inference to output the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment. The planned irrigation duration or irrigation volume can be output daily to form a multi-day irrigation plan, or the multi-day plan can be broken down into a sequence of control instructions for successive irrigation events to adapt to the execution methods of different irrigation facilities.

[0068] Real-time root zone soil moisture content is obtained by mapping multi-depth soil moisture content through root zone depth. Specifically, the upper and lower boundary depths of the root zone are first determined to define the root zone depth range. Then, soil moisture contents at each depth falling within the root zone range are selected from the multi-depth soil moisture contents as a candidate set for the root zone. Finally, the candidate set is fused according to a preset depth decay function to obtain the real-time root zone soil moisture content. The depth decay function reflects the non-uniformity of root water absorption contribution with depth, ensuring that the moisture content near the upper part of the root zone or the dense root layer contributes more to the real-time root zone soil moisture content, while the contribution from the lower part of the root zone is relatively reduced. This makes the root zone index closer to the effective water content characterization required for irrigation decisions.

[0069] The input variables of the fuzzy rule base include at least the predicted water demand, real-time root zone soil moisture content, and growth stage index, each with at least three levels of fuzzy linguistic sets. Specifically, the fuzzy linguistic set for the predicted water demand is used to characterize the discrete state of future water demand levels; the fuzzy linguistic set for the real-time root zone soil moisture content is used to characterize the discrete gradation of root zone water status; and the fuzzy linguistic set for the growth stage index is used to characterize the differences in water response and irrigation demand when the crop is at different growth stages. Triangular or Gaussian membership functions are used to fuzzify the above input variables to obtain the membership degree distribution of each input variable on the corresponding fuzzy linguistic set, providing a basis for subsequent rule triggering intensity calculations.

[0070] The output variable of the fuzzy rule base is the planned irrigation duration or irrigation volume. The trigger strength of each rule is calculated based on the membership distribution of the input variables, and the rule outputs are weighted and aggregated using dynamic rule weights. Specifically, for each rule, its dynamic rule weight and trigger strength are used to determine its contribution. This contribution modulates the rule's proportion in the overall output, thus achieving the control characteristic that "the rule contribution can adaptively change with the operating effect under the same input conditions." Subsequently, the center-of-gravity method is used to defuzzify the weighted aggregation result, obtaining continuous output values. These continuous output values ​​are used as the planned irrigation duration or planned irrigation volume to directly drive valve opening control, irrigation duration control, or irrigation volume control.

[0071] The fuzzy rule base comprises a set of rules covering multiple combinations of water demand levels, soil moisture states, and growth stages, with no fewer than twenty-seven rules. This ensures the rule base covers the main combination space formed by at least three levels of fuzzy language sets for each of the three input variables. Each rule is associated with a dynamic weight, which reflects the long-term effectiveness and reliability of the rule under the current plot, season, and crop stage. When the dynamic weight of a rule falls below a preset lower limit, a rule protection mechanism is implemented: firstly, the dynamic weight is frozen to prevent it from continuing to decrease with short-term deviations, thus avoiding weight oscillations; secondly, the rule is temporarily disabled, preventing it from participating in the weighted inference of the current cycle. This reduces the interference of unreliable rules on the planned irrigation duration or irrigation volume output, thereby improving the stability and controllability of the fuzzy decision output.

[0072] After each irrigation cycle, a post-irrigation assessment process is initiated. Rainfall and soil moisture content at multiple depths within the root zone are continuously collected within a preset assessment time window to calculate changes in root zone soil moisture storage and determine actual water consumption. The preset assessment time window covers the main stages of post-irrigation water infiltration and redistribution, ensuring the assessment results reflect the true contribution of this irrigation to effective root zone replenishment and providing a reliable basis for subsequent dynamic weight updates.

[0073] The actual water consumption is calculated within a preset assessment time window. Specifically, the irrigation volume corresponding to the irrigation and the precipitation within the assessment time window are obtained, and the root zone soil moisture storage at the start and end of the assessment time window is determined, thus obtaining the change in root zone soil moisture storage. Root zone soil moisture storage can be obtained by converting and integrating the volumetric water content of soil at each depth within the root zone depth range with the corresponding soil layer thickness, or by converting the representative value of root zone soil water content with the root zone thickness. Subsequently, the change in root zone soil moisture storage is subtracted from the sum of irrigation volume and precipitation to obtain the actual water consumption. This allows the actual water consumption to simultaneously represent the comprehensive results of crop transpiration consumption, surface evaporation consumption, and possible deep seepage losses, thereby forming a comparable evaluation standard with the predicted water demand.

[0074] The dynamic weights of rules in the fuzzy rule base are updated based on the deviation between actual water consumption and predicted water demand. Specifically, the predicted water demand for one to three days corresponding to this irrigation decision is compared with the actual water consumption obtained within the evaluation time window to generate a deviation. This deviation is then used to update the dynamic weights of the rules participating in this inference: when the deviation indicates that actual water consumption is significantly higher than the predicted water demand, the dynamic weights of rules related to high irrigation volume or long irrigation duration are reduced to suppress over-irrigation tendencies; when the deviation indicates that actual water consumption is significantly lower than the predicted water demand, the dynamic weights of rules related to insufficient water replenishment correction are increased to enhance the water replenishment intensity of subsequent irrigation plans. The updated dynamic weights of the rules, together with the corresponding rule triggering intensity, determine the rule contribution in subsequent weighted inference, thereby achieving adaptive correction of planned irrigation duration or irrigation volume.

[0075] Within a pre-defined evaluation time window after each irrigation, a root zone water retention potential is constructed to characterize the effective retention degree of the irrigation within the root zone. Specifically, firstly, the root zone depth range is determined, and soil moisture content at multiple depths within the root zone is selected as the root zone observation set. Then, the increase in root zone soil moisture storage from the end of irrigation to the end of the evaluation time window is calculated. The root zone retention ratio is then characterized by the ratio of the increase in root zone soil moisture storage to the sum of the irrigation amount and precipitation. The root zone retention ratio is monotonically mapped and normalized to obtain the root zone water retention potential, making it comparable and controllable within a pre-defined interval.

[0076] Root zone water retention potential is used as an additional input for fuzzy inference or as a correction factor for planned irrigation duration or irrigation volume in control decisions. Specifically, when the root zone water retention potential indicates a low effective retention capacity in the root zone, it is determined that there is a risk of infiltration penetration or difficulty in retaining water in the root zone. Therefore, the planned irrigation duration or planned irrigation volume is reduced, or a single planned irrigation duration is divided into multiple execution segments to reduce the infiltration intensity per unit time, making the water replenishment more aligned with the water absorption process in the root zone. When the root zone water retention potential is high, the planned output is allowed to more fully respond to the water demand forecast, thereby improving water use efficiency while controlling the risk of excessive infiltration.

[0077] Within a pre-defined evaluation time window after each irrigation, a penetration sequence deviation is constructed to characterize whether the penetration response of water from shallow to deep conforms to the expected infiltration sequence. Specifically, for soil moisture content at multiple depths, the response time at which the soil moisture content at each depth reaches a pre-defined increase threshold from the end of irrigation is determined, and a reference penetration sequence with increasing response times from shallow to deep is established. Subsequently, the measured response time sequence is compared with the reference penetration sequence, and a deviation is generated based on the number of reverse occurrences and the degree of abrupt change in the difference between response times at adjacent depths. The deviation is then normalized to obtain the penetration sequence deviation, enabling stable comparisons across plots and stages.

[0078] When a certain depth does not reach the preset increase threshold within the preset evaluation time window, a boundary processing strategy is adopted to avoid distortion in the deviation calculation: the end time of the preset evaluation time window is taken as the response time of that depth, or the depth is marked as non-responsive and included in the deviation according to the preset penalty. Through this strategy, it is possible to distinguish between the different situations of "no obvious penetration response" and "penetration response occurs but the time sequence is reversed", ensuring that the deviation of the penetration sequence law has a consistent characterization of the over-permeability risk.

[0079] The deviation of the penetration order law serves as a constraint on over-infiltration risk and participates in fuzzy reasoning and dynamic rule weight updates. Specifically, when the deviation of the penetration order law exceeds a preset threshold, an over-infiltration or abnormal penetration risk is identified. The dynamic weight of the rules corresponding to high irrigation volumes or long irrigation durations is reduced, and protection strategies are triggered, including reducing valve opening or shortening the duration of a single irrigation session, to reduce the infiltration driving force and limit the tendency for deep penetration. When the deviation of the penetration order law is within a safe range, an irrigation plan is generated based on the predicted water demand and root zone status according to conventional strategies, thereby improving irrigation safety and controllability while ensuring crop water demand.

[0080] Example 2:

[0081] Please see Figure 2 This exemplary short-term crop irrigation forecasting system based on fuzzy control, used to implement the aforementioned short-term crop irrigation forecasting method based on fuzzy control, includes: a processor, a memory, and a communication interface. The memory stores a program executable by the processor, and the processor executes the program to implement the following functional units:

[0082] Multi-source acquisition module: Collects multi-dimensional time-series data in the field, including at least meteorological data, soil moisture content at multiple depths and crop growth stage data, and obtains feedback on irrigation flow or irrigation volume;

[0083] Preprocessing module: performs time alignment, anomaly removal, missing data imputation, and normalization on time-series data;

[0084] Short-term forecasting module: Input the processed time series data into the time series forecasting model with a preset historical time window, and output the crop water requirement forecast value and confidence level for one to three days. The confidence level is obtained by monotonic mapping and normalization based on the historical verification error of the time series forecasting model.

[0085] Fuzzy decision module: It fuzzifies the water demand forecast, real-time root zone soil moisture content and growth stage index, performs weighted reasoning based on a fuzzy rule base with dynamic weights and defuzzifies it, and outputs the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment.

[0086] Feedback self-adjustment module: Within the preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in soil moisture storage in the root zone from the sum of irrigation volume and precipitation. The dynamic weights of the rules in the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, which is used to adaptively correct the planned irrigation duration or irrigation volume in the future.

[0087] It should be noted that the short-term crop irrigation forecasting system based on fuzzy control provided in the above embodiments and the short-term crop irrigation forecasting method based on fuzzy control provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the short-term crop irrigation forecasting system based on fuzzy control provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0088] 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 technical scope 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.

Claims

1. A short-term crop irrigation forecasting method based on fuzzy control, characterized in that, include: Collect multi-dimensional time-series data in the field, including at least meteorological data, soil moisture content at multiple depths and crop growth stage data, and obtain feedback on irrigation flow or irrigation volume. Perform time alignment, outlier removal, missing data imputation, and normalization on time-series data; The processed time series data is input into a time series prediction model with a preset historical time window, and the predicted crop water requirement and confidence level for one to three days are output. The confidence level is obtained by monotonic mapping and normalization based on the historical verification error of the time series prediction model. The predicted water demand, real-time root zone soil moisture content, and growth stage index are fuzzified. Weighted reasoning and defuzzification are performed based on a fuzzy rule base with dynamic weights and rules. The output is the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment. Within a preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in soil moisture storage in the root zone from the sum of irrigation volume and precipitation. The dynamic weights of the rules in the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, which is used to adaptively correct the planned irrigation duration or irrigation volume for subsequent irrigation.

2. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, Meteorological data include temperature, humidity, wind speed, radiation, and precipitation; multi-depth soil moisture content includes at least the volumetric moisture content of the shallow, middle, and deep root zones; crop growth stage data includes growth stage indicators or growth stage indices; irrigation flow or irrigation volume feedback includes at least one of the following: irrigation start time, duration, valve open / closed status, and valve opening degree.

3. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, Anomaly removal includes joint detection of mutations based on thresholds of changes in adjacent time periods and outlier detection based on historical distribution; missing information filling includes completion based on temporal imputation or imputation based on similar day nearest neighbors; Normalization is based on the dimensional unification of sliding time window statistics, which allows meteorological data, soil moisture content, and growth stage indices to be used as inputs for time series forecasting models.

4. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, The real-time root zone soil moisture content is obtained by mapping the soil moisture content at multiple depths through the root zone depth. The root zone depth mapping includes: determining the upper and lower boundary depths of the root zone; and fusing the moisture content at each depth falling within the root zone range according to a preset depth attenuation function to obtain the real-time root zone soil moisture content.

5. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, The time-series prediction model adopts a bidirectional time-series network with an attention mechanism. The input is a sequence of meteorological data, a sequence of soil moisture content at multiple depths, and a sequence of growth stage indexes within a preset historical time window. The output is the predicted crop water requirement for one to three days respectively. The confidence level is obtained by monotonically mapping and normalizing the historical verification error obtained by the time-series prediction model during the training and verification phase, and is set as a risk suppression factor for subsequent fuzzy inference.

6. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, The input variables of the fuzzy rule base include at least the predicted water demand, the real-time root zone soil moisture content, and the growth stage index, and each has a fuzzy language set of no less than three levels. The input variables are fuzzified using triangular membership functions or Gaussian membership functions. The output variables are the planned irrigation duration or irrigation volume, and the continuous output values ​​are obtained by defuzzifying using the centroid method.

7. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, The fuzzy rule base includes a set of rules covering multiple water demand levels, multiple soil moisture states, and multiple growth stage combinations, with no fewer than twenty-seven rules. Each rule is associated with a dynamic rule weight, and the dynamic rule weight, together with the corresponding rule trigger strength, determines the contribution of the rule during weighted inference. When the dynamic rule weight is lower than a preset lower limit, the dynamic rule weight is frozen or the rule is temporarily disabled.

8. The short-term crop irrigation forecasting method based on fuzzy control according to claim 1, characterized in that, Within a preset evaluation time window after each irrigation, the root zone water retention potential is constructed. The determination of the root zone water retention potential includes: determining the root zone depth range and selecting soil moisture content at multiple depths within the root zone; calculating the increase in root zone soil moisture storage from the end of irrigation to the end of the evaluation time window, and characterizing the root zone retention ratio by the ratio of the increase in root zone soil moisture storage to the sum of irrigation amount and precipitation; obtaining the root zone water retention potential by monotonic mapping and normalization of the retention ratio; and using the root zone water retention potential as an additional input for fuzzy inference or as a correction factor for the planned irrigation duration or irrigation amount. When the root zone water retention potential is lower than a preset threshold, the planned irrigation duration is reduced or the planned irrigation duration is divided into multiple segments for execution.

9. A short-term crop irrigation forecasting method based on fuzzy control according to claim 8, characterized in that, Within a preset evaluation time window after each irrigation, a penetration order deviation is constructed. The determination of the penetration order deviation includes: for soil moisture content at multiple depths, determining the response time at which the soil moisture content at each depth reaches a preset increase threshold from the end of irrigation; establishing a reference penetration order with increasing response times from shallow to deep; comparing the measured response time sequence with the reference penetration order, generating a deviation degree based on the number of reverse occurrences and the degree of abrupt change in the response time difference between adjacent depths, and normalizing it to obtain the penetration order deviation; wherein, when a certain depth does not reach the preset increase threshold within the preset evaluation time window, the end time of the preset evaluation time window is taken as the response time of that depth, or the depth is marked as non-responsive and included in the deviation degree according to a preset penalty term; the penetration order deviation is used as an over-infiltration risk constraint in fuzzy reasoning and rule dynamic weight update, wherein when the penetration order deviation exceeds the preset threshold, the rule dynamic weight corresponding to high irrigation volume or long irrigation duration is reduced and a protection strategy of reducing valve opening or shortening single irrigation duration is triggered.

10. A short-term crop irrigation forecasting system based on fuzzy control, used to implement the short-term crop irrigation forecasting method based on fuzzy control as described in any one of claims 1-9, characterized in that, include: The processor, memory, and communication interface are included. The memory stores programs that can be executed by the processor, and the processor executes the programs to implement the following functional units: Multi-source acquisition module: Collects multi-dimensional time-series data in the field, including at least meteorological data, soil moisture content at multiple depths and crop growth stage data, and obtains feedback on irrigation flow or irrigation volume; Preprocessing module: performs time alignment, anomaly removal, missing data imputation, and normalization on time-series data; Short-term forecasting module: Input the processed time series data into the time series forecasting model with a preset historical time window, and output the crop water requirement forecast value and confidence level for one to three days. The confidence level is obtained by monotonic mapping and normalization based on the historical verification error of the time series forecasting model. Fuzzy decision module: It fuzzifies the water demand forecast, real-time root zone soil moisture content and growth stage index, performs weighted reasoning based on a fuzzy rule base with dynamic weights and defuzzifies it, and outputs the planned irrigation duration or irrigation volume for one to three days to drive the irrigation equipment. Feedback self-adjustment module: Within the preset evaluation time window after irrigation, the actual water consumption is obtained by subtracting the change in soil moisture storage in the root zone from the sum of irrigation volume and precipitation. The dynamic weights of the rules in the fuzzy rule base are updated based on the deviation between the actual water consumption and the predicted water demand, which is used to adaptively correct the planned irrigation duration or irrigation volume in the future.

Images