Cotton irrigation optimization method and system

Abstract

The invention discloses a cotton irrigation optimization method and system, and relates to the technical field of automatic control. The method comprises the following steps: step 1, forming an original cotton field environment information set through multi-dimensional sensing nodes arranged in a target cotton field; 2, the central control terminal carries out time sequence synchronization processing on the original cotton field environment information set, a historical cotton field yield matrix is introduced into a reorganized cotton field environment information matrix, and a cotton field cooperation data body is generated; 3, calling the vortex pulse weaving probability calculation model in the central control terminal, and outputting a cotton field dynamic moisture demand prediction sequence; 4, the central control terminal compares the current available water resource quantity according to the cotton field dynamic water demand prediction sequence, and cotton irrigation optimization is completed. The irrigation scientificity, the crop yield stability and the water resource utilization efficiency are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of automatic control technology, and in particular to a cotton irrigation optimization method and system. Background Art

[0002] In the current agricultural irrigation management system, precision irrigation has gradually replaced traditional empirical irrigation and become one of the primary means of improving water resource utilization and crop yield stability. In particular, in cotton cultivation, the scientific nature of irrigation decisions is directly related to the water physiological status of cotton plants and yield formation. However, based on existing practical applications and publicly available technical means, existing irrigation optimization technologies still have many limitations in terms of the granularity of information perception, the dynamic nature of model calculations, the ability to express spatiotemporal structures, and the adaptability of the final scheduling strategy. These limitations do not yet fully meet the needs for efficient management of water regulation in large-scale cotton fields across different climate zones, different tillage systems, and different time periods.

[0003] Currently, most common irrigation technologies are based on threshold-triggered control, which automatically triggers irrigation when a certain critical value of soil moisture or environmental indicators is set. This method is simple to implement and has low deployment costs, but its fundamental problem is that its ability to describe the dynamic water needs of crops is very limited. Such systems often only perceive the moisture content of a single layer of soil, ignoring the combined effects of crop transpiration, climatic factors, and soil thermal dynamics. Therefore, their triggering mechanism cannot reflect the actual degree of crop water stress. In addition, fixed thresholds are not adaptable in time and space, and show large deviations when faced with different growth periods, different plot conditions, and rapidly changing climate environments, which can easily lead to over-irrigation or under-irrigation. Summary of the Invention

[0004] In view of this, the present invention provides a cotton irrigation optimization method and system. By constructing a reorganized cotton field environmental information matrix and a cotton field collaborative data body, multiple vortex levels and pulse winding mechanisms are used to dynamically calculate the water demand prediction sequence, and irrigation time windows and water scheduling plans are generated in combination with available water resources. Accurate modeling and efficient regulation of the moisture status of the cotton field are achieved, effectively improving the scientific nature of irrigation, crop yield stability and water resource utilization efficiency.

[0005] The technical solution adopted in the present invention is as follows:

[0006] A cotton irrigation optimization method, comprising:

[0007] Step 1: Through the multi-dimensional sensing nodes deployed in the target cotton field, soil moisture content, soil temperature, crop leaf moisture, wind speed and sunshine intensity are continuously acquired and uploaded to the central control terminal in real time to form the original cotton field environmental information set;

[0008] Step 2: The central control terminal performs time-series synchronization processing on the original cotton field environmental information set, using timestamp alignment to eliminate cross-node sampling deviations to obtain time-series synchronized data; the time-series synchronized data is partitioned and marked based on the plot number to obtain the reorganized cotton field environmental information matrix; the historical cotton field yield matrix is ​​introduced into the reorganized cotton field environmental information matrix to generate a cotton field collaborative data volume;

[0009] Step 3: Invoke the vortex pulse weaving probability calculation model in the central control terminal, perform decision calculations on the cotton field collaborative data volume, and output a dynamic water demand prediction sequence for the cotton field;

[0010] Step 4: The central control terminal compares the current available water resources based on the dynamic water demand prediction sequence of the cotton field, uses the partition priority queue method to generate an irrigation time window list and a corresponding irrigation water volume list to complete the cotton irrigation optimization.

[0011] Furthermore, in step 2, the time series synchronization data is partitioned and marked based on the plot number to obtain the reorganized cotton field environment information matrix. The process includes: the central control terminal reads the record entries one by one from the time series synchronization data, traverses the field set corresponding to each record entry, identifies and extracts the plot number field therein; the central control terminal generates a plot mapping table based on all the obtained plot number fields, which uses the plot number as the primary key and the sequentially increasing partition index number as the value, and performs a uniqueness check during the writing process to avoid the same plot number being repeatedly mapped to different partition index numbers; the central control terminal calls the plot mapping table and performs the following processing on each record entry in the time series synchronization data: reads the plot number field of the record entry; queries the plot mapping table, and obtains the corresponding partition Index number; create a new partition mark field in the record entry and write the partition index number; store the record entry after writing the partition mark field into the buffer area to be reorganized; the central control terminal partitions and archives the record entries in the buffer area to be reorganized according to the partition mark field, and stores all record entries under the same partition mark field value in the same partition storage area in sequence; keep the original time sequence unchanged during the storage process; the central control terminal constructs a row vector in each partition storage area according to the preset field order; arranges all row vectors in the same partition storage area in ascending order according to the sampling timestamp field; writes the sorted row vectors in each partition storage area into the matrix row cache in sequence to form partition row segments; merges all partition row segments in ascending order of the partition index number to generate the reorganized cotton field environmental information matrix.

[0012] Furthermore, in step 2, the process of generating the cotton field collaborative data body includes: the central control terminal adds a time pairing field to each record in the historical cotton field yield matrix, converts the harvest year field into a unified timestamp of the last day of the year, and writes the time pairing field to form a time dimension correspondence with the sampling timestamp field in the reorganized cotton field environmental information matrix; the reorganized cotton field environmental information matrix is ​​indexed in ascending order according to the plot number field and the sampling timestamp field; the historical cotton field yield matrix is ​​indexed in ascending order according to the plot number field and the time pairing field, and an inner join operation is performed on the reorganized cotton field environmental information matrix and the historical cotton field yield matrix to obtain the cotton field collaborative data body.

[0013] Furthermore, step 3 specifically includes:

[0014] Step 3.1: Construct multiple vortex hierarchies and map the cotton field collaborative data volume into the inner vortex layer and the outer vortex layer in the spatial dimension and the temporal dimension respectively;

[0015] Step 3.2: Trigger a pulsed weaving operation to perform complementary winding of the node weights in the inner and outer vortex layers in an alternating order;

[0016] Step 3.3: After each weaving cycle, select the surviving weighted path based on the probability drift rule;

[0017] Step 3.4: Continue iterating until the preset convergence threshold is met to output the dynamic water demand prediction sequence of the cotton field.

[0018] Furthermore, step 3.1 specifically includes: the central control terminal reads the cotton field collaborative data body, and extracts the plot number field, sampling timestamp field and partition mark field for each row vector; wherein, the plot number field and the partition mark field together constitute the spatial dimension index key, and the sampling timestamp field constitutes the time dimension index key; the central control terminal creates a new multiple vortex hierarchical structure buffer, and presets the hierarchical index numbers of the first inner vortex layer and the first outer vortex layer in the buffer, which are respectively recorded as inner vortex index number one and outer vortex index number one; at the same time, a hierarchical mapping table is created, with the hierarchical index number as the primary key and the vortex hierarchical depth number as the value; for each row vector in the cotton field collaborative data body, the central control terminal performs spatial dimension mapping to form an inner vortex node, and then performs time dimension mapping on the same row vector to form an outer vortex node; the buffer zone of the inner vortex layer node and the outer vortex layer node are mapped. For the nodes of the vortex layer node buffer, the central control terminal establishes a one-to-one corresponding node connection relationship based on the original row number of the row vector in the cotton field collaborative data body; the corresponding relationship is written into the vortex connection table; when the number of nodes in the inner vortex layer node buffer or the outer vortex layer node buffer reaches the preset level expansion threshold, the central control terminal creates the next inner vortex layer and the next outer vortex layer in the multiple vortex level construction buffer, and assigns a new level index number to it; then returns to execute spatial dimension mapping and time dimension mapping until the cotton field collaborative data body is fully mapped; the central control terminal performs an integrity check on the multiple vortex level construction buffer, and after confirming that the inner vortex layer node buffer, the outer vortex layer node buffer and the vortex connection table are fully recorded, all data is written into the vortex level storage area, and the level solidification timestamp is recorded to complete the construction of the multiple vortex level.

[0019] Furthermore, the process of the central control terminal performing spatial dimension mapping specifically includes: for each row vector in the cotton field collaborative data body, the central control terminal performs spatial dimension mapping according to the spatial dimension index key, specifically including: if the corresponding inner vortex index number does not exist in the hierarchical mapping table, the central control terminal assigns a new inner vortex index number to the spatial dimension index key, and registers it in the hierarchical mapping table; the spatial dimension index key of the row vector is bound to the corresponding inner vortex index number and written into the inner vortex layer node buffer to form an inner vortex node; the process of the central control terminal performing time dimension mapping specifically includes: the central control terminal performs time dimension mapping on the time dimension index key of the same row vector, specifically including: if the corresponding outer vortex index number does not exist in the hierarchical mapping table, the central control terminal assigns a new outer vortex index number to the time dimension index key, and registers it in the hierarchical mapping table; the time dimension index key of the row vector is bound to the corresponding outer vortex index number and written into the outer vortex layer node buffer to form an outer vortex node.

[0020] Furthermore, step 3.2 specifically includes: the central control terminal generates an initial pulse signal, and simultaneously sends an initial pulse signal to the inner vortex layer and the outer vortex layer in the constructed multiple vortex levels, activates all nodes in the two vortex layers, and enters the waiting-for-weaving state; the central control terminal adds a node weight field to the nodes in each vortex layer, and assigns an initial node weight to each node; the central control terminal repeats the following operations according to the preset complementary winding order, starting from the inner vortex layer, until a preset number of complementary winding cycles are completed: each node in the inner vortex layer selects a node in the corresponding outer vortex layer with a connection relationship established, reads the node weight of the other party, and performs a complementary winding operation with its own node weight, and updates its own node weight; each node in the outer vortex layer selects a node in the corresponding inner vortex layer with a connection relationship established, reads the node weight of the other party, and performs a complementary winding operation with its own node weight, and updates its own node weight.

[0021] Furthermore, step 3.3 specifically includes: after each weaving cycle, the central control terminal extracts the node weights of all nodes in the inner vortex layer and the outer vortex layer in turn, and arranges them in sequence according to the space and time indexes corresponding to the nodes to form a node weight drift path matrix; the central control terminal inputs the path entropy vector into the fuzzy membership evaluation function to obtain the fuzzy membership value; at the same time, the drift probability value is calculated based on the cumulative amplitude of the node weight change; the fuzzy membership value and the drift probability value are linearly coupled to obtain the fuzzy entropy drift index; the central control terminal presets two-level probability drift rule thresholds of upper threshold and lower threshold: when the fuzzy entropy drift index is greater than or equal to the upper threshold, the corresponding node weight path sequence is directly marked as a surviving weight path sequence; when the fuzzy entropy drift index is between the upper threshold and the lower threshold, the secondary judgment process is entered, and the local time window stability test is called. If the test passes, it is marked as a weight path sequence to be survived; when the fuzzy entropy drift index is less than the lower threshold, the corresponding node weight path sequence is directly deleted.

[0022] Furthermore, step 3.4 specifically includes: the central control terminal extracts all node weights in the survival weight path sequence and the to-be-survived weight path sequence, and rearranges the node weights according to the spatial dimension index key and the time dimension index key; then calls the smoothing processing module to perform time series smoothing based on cubic spline interpolation to complete the generation of the cotton field dynamic water demand prediction sequence.

[0023] A cotton irrigation optimization system comprises: multidimensional sensing nodes and a central control terminal; the multidimensional sensing nodes are arranged in a target cotton field, continuously acquiring soil moisture content, soil temperature, crop leaf surface humidity, wind speed and sunshine intensity, and uploading the information to the central control terminal in real time to form an original cotton field environmental information set; the central control terminal performs time series synchronization processing on the original cotton field environmental information set, adopts a timestamp alignment method to eliminate cross-node sampling deviation, and obtains time series synchronization data; the time series synchronization data is partitioned and marked based on the plot number to obtain a reorganized cotton field environmental information matrix; a historical cotton field yield matrix is ​​introduced into the reorganized cotton field environmental information matrix to generate a cotton field collaborative data body; an eddy pulse weaving probability calculation model is called to perform decision calculation on the cotton field collaborative data body and output a cotton field dynamic water demand prediction sequence; based on the cotton field dynamic water demand prediction sequence, the current available water resources are compared, and a partition priority queue method is adopted to generate an irrigation time window list and a corresponding irrigation water amount list to complete cotton irrigation optimization.

[0024] By adopting the above technical solution, the present invention has the following beneficial effects: through the continuous collection of soil moisture content, soil temperature, crop leaf surface humidity, wind speed and sunshine intensity by multi-dimensional sensing nodes, a full-factor environmental information system covering water, heat, transpiration and climate conditions is constructed, significantly improving the basic data accuracy of irrigation decision-making. Secondly, the central control terminal uses timestamp alignment to complete the time series synchronization processing of asynchronous data, and combines the plot number and partition mark to achieve highly consistent spatial classification, thereby generating a structurally stable reorganized cotton field environmental information matrix, solving the problems of data dispersion and inconsistent structure in existing irrigation systems. Furthermore, the system introduces the historical cotton field yield matrix and integrates it with the environmental information matrix to form a cotton field collaborative data body, enabling the model to combine short-term environmental dynamics with long-term yield feedback for joint modeling, thereby improving the causal rationality of the prediction results. The present invention also constructs a space-time coupled graph structure based on multiple vortex levels, and iteratively updates the node weights through a complementary winding mechanism driven by pulses. Under the guidance of the fuzzy entropy drift rule, stable and information-valuable weight paths are screened to achieve accurate identification of water demand trends. Finally, the weighted results were smoothed by cubic spline interpolation to obtain a continuous, controllable, and physically meaningful cotton field dynamic water demand prediction sequence, which provided direct support for the subsequent generation of irrigation time window lists and irrigation water volume lists. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a schematic diagram of a cotton irrigation optimization method according to an embodiment of the present invention;

[0026] Figure 2 2. Schematic diagram of analysis of fuzzy entropy drift index and survival weight path screening effect in an embodiment of the present invention;

[0027] Figure 3 3. This is a schematic diagram comparing the accuracy of the cotton field dynamic water demand prediction sequence before and after cubic spline interpolation smoothing processing in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] All features disclosed in this specification, or all steps in the disclosed methods or processes, except mutually exclusive features and / or steps, can be combined in any manner.

[0029] Any feature disclosed in this specification (including any appended claims and abstract), unless otherwise stated, may be replaced by other equivalent or similar features. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0030] refer to Figure 1 : A cotton irrigation optimization method, the method comprising:

[0031] Step 1: Through the multi-dimensional sensing nodes deployed in the target cotton field, soil moisture content, soil temperature, crop leaf moisture, wind speed and sunshine intensity are continuously acquired and uploaded to the central control terminal in real time to form the original cotton field environmental information set;

[0032] Step 2: The central control terminal performs time-series synchronization processing on the original cotton field environmental information set, using timestamp alignment to eliminate cross-node sampling deviations to obtain time-series synchronized data; the time-series synchronized data is partitioned and marked based on the plot number to obtain the reorganized cotton field environmental information matrix; the historical cotton field yield matrix is ​​introduced into the reorganized cotton field environmental information matrix to generate a cotton field collaborative data volume;

[0033] Step 3: Invoke the vortex pulse weaving probability calculation model in the central control terminal, perform decision calculations on the cotton field collaborative data volume, and output a dynamic water demand prediction sequence for the cotton field;

[0034] Step 4: The central control terminal compares the current available water resources based on the dynamic water demand prediction sequence of the cotton field, uses the partition priority queue method to generate an irrigation time window list and a corresponding irrigation water volume list to complete the cotton irrigation optimization.

[0035] In this method, multidimensional sensing nodes perform fine-grained state acquisition of the target cotton field. Their deployment principle stems from the sufficient sampling assumption in information theory: as long as sensors are spatially distributed at a certain density and maintain a fixed sampling frequency over time, a continuous environmental field can be approximated using limited data. Nodes select soil moisture, soil temperature, crop leaf wetness, wind speed, and sunlight intensity as core observation indicators because these five variables collectively determine the instantaneous fluxes of cotton plant transpiration and soil evaporation, and moisture flux is the direct driver of irrigation decisions. Each node maintains a microsecond-level time reference locally via a crystal oscillator, but random delays introduced by wireless transmission links can still cause data from different nodes to arrive at the central control terminal at different times. To ensure physical consistency in the time axis of subsequent computational views, the central control terminal first timestamp-aligns the raw cotton field environmental information set. This principle can be viewed as a unified sequence mapping: the system establishes a reference axis with millisecond resolution, projects the raw sampling points uploaded by each node onto the reference axis, and fills in missing node dimensions at the same time using linear interpolation from adjacent time points, thereby generating time-synchronized data. The interpolation process essentially treats the measurement points as discrete samples on a space-time binary function and completes the reconstruction through the simplest linear basis function, which not only avoids oscillations caused by high-order interpolation but also allows for continued iterative refinement within the subsequent model.

[0036] After obtaining time-series synchronized data, the central control terminal needs to spatially partition and mark the data. The reason for using plot numbers rather than direct latitude and longitude coordinates for partitioning is that field operations are typically managed using plots as the smallest unit. Irrigation pipe networks, pump station valves, and machinery routes are all located around plot boundaries. Directly slicing by coordinates would result in discrepancies between the generated plots and the actual irrigation control units, making implementation difficult. The core principle of the partition marking process is to establish an injective mapping table, where each plot number corresponds to only one partition index. This injective property ensures that subsequent matrix operations avoid ambiguity, where the same row belongs to multiple partitions. A uniqueness check is performed when writing to the mapping table, essentially a key uniqueness constraint in set theory, to avoid mapping conflicts. After marking, the system uses the partition index and sampling timestamp as row pointers, and the corresponding column pointers maintain the five core indicators. This resulting compiled cotton field environmental information matrix satisfies the canonical matrix form, with isomorphic rows and isodimensional columns, making it suitable for high-dimensional linear algebra operations.

[0037] However, real-time environmental indicators alone cannot capture the cumulative response of cotton plants to water stress. To enable the model to understand the historical yield consequences of similar environmental trajectories, the central control terminal imports the historical cotton field yield matrix and connects it to the reorganized cotton field environmental information matrix through a time-matching field. The inner join operation here utilizes the database concept of equivalent joins: row vectors can only be merged if both the plot number and the time dimension index are matched. Yield data is aggregated by year, while environmental data is presented at the second level. The two time scales differ significantly. The end of the year is selected as a unified timestamp to ensure that the annual yield is backward closed to the annual environmental segment on the time axis. This closure allows the cotton field collaborative data volume to simultaneously encompass micro-environmental changes and macro-result feedback, providing a traceable causal clue for subsequent machine reasoning.

[0038] The vortex pulse weaving probabilistic calculation model is the algorithmic core of this method, deriving its concept from the abstraction of complex network coupling dynamics. The model decomposes the cotton field collaborative data volume into multiple vortex levels, mapping the spatial dimension into the inner vortex layer and the temporal dimension into the outer vortex layer. The word "vortex" implies a rotational structure—in network topology, nodes in the inner and outer vortex layers form a circular reference chain through the vortex connection table. This circular structure allows weights to automatically propagate back and forth during iteration, simulating a closed feedback loop between field environmental factors and historical yields. The weaving process incorporates pulse signals, recognizing that biophysical systems often exhibit pulsed energy release. Weight complementary weaving is a bidirectional exchange of weights driven by pulses. The weaving operation adheres to the principle of complementarity: the increment in the weight of an inner vortex node is equal to a function mapping of the outer vortex node weight, and vice versa. After multiple iterations, the overall network weight tends to equilibrium. At the end of each weaving cycle, the system analyzes the node weight drift path, essentially using information entropy and fuzzy membership to evaluate the path's contribution to the overall output. Paths with high entropy and high membership represent information-rich and stable paths, distinguishing them from noisy paths with low entropy and low membership. The probability drift rule forms a dual-threshold strategy based on upper and lower thresholds. In machine learning, this is equivalent to using a soft-gated attenuation mechanism with hysteresis to prune weights, preserving informative associations while eliminating excess noise. Surviving weighted paths and pending survival weighted paths are retained until the next iteration, triggering shutdown at the preset convergence threshold. The output sequence of dynamic cotton field water demand predictions represents the model's estimate of water demand for each subarea in the future.

[0039] With a dynamic cotton field water demand forecast sequence, resource control can begin by comparing it with the available water resources maintained in real time by the central control terminal. The core principle here is a combination of queuing theory and heuristic scheduling. The system calculates a water shortage risk index and crop growth stage weight for each zone. These two are weighted to form a priority score, which is then placed into a zone priority queue. The zone with the most urgent water needs always remains at the head of the queue. When generating the irrigation time window list and irrigation water volume list, the system examines the pump station load and pipe network pressure curves for the previous period. Using peak flow as a hard constraint, a delay-tolerant strategy is used to shift low-priority zones to a later peak. Peak shifting is not a simple delay; it integrates multiple factors, such as the low nighttime transpiration of cotton plants and low electricity prices at the pump station, to ensure the physiological safety zone of the crop while also improving energy efficiency. After a complete scheduling cycle is completed, the list is sent to the pump control unit and valve control unit at the execution layer, which translates the algorithm output into physical irrigation actions.

[0040] Furthermore, during the implementation of cotton irrigation optimization, the central control terminal needs to convert the time-series synchronization data from an unordered streaming collection into a reorganized cotton field environmental information matrix that can be directly called by subsequent algorithms. The core of this conversion is to complete the entire process of partition marking based on the plot number field. Its principle can be understood as a bottom-up data hierarchical return: First, the central control terminal reads the record entries in the time-series synchronization data one by one according to the order of arrival, and locks the plot number field when traversing the corresponding field set, because this field naturally corresponds to the smallest physical unit of field management and is also a direct index for the subsequent partition control valve action. In order to ensure that the same plot number always falls into a unique logical partition, the system establishes a plot mapping table, and uses the plot number field as the primary key and the sequentially increasing partition index number as the value. The construction of the mapping table is an online incremental process: whenever the system discovers a new plot number, it immediately assigns the next partition index number; before the record entry is written to the mapping table, the system confirms through a uniqueness check that the plot number has not been previously registered, avoiding conflicts where the same plot number is mapped to different partition index numbers, thereby maintaining the single-value property of primary key in the sense of set theory.

[0041] After completing the dynamic maintenance of the mapping table, the central control terminal calls it to perform real-time annotation of the time-series synchronization data. Each time a record entry is processed, the block number field of the record entry is read, and the corresponding partition index number is immediately obtained through a hash query. A new partition tag field is then created in the record entry and the partition index number is written. The marked record entries are cached in the pending compilation buffer, which is a linked list structure in memory. It maintains the write order and provides an entry for batch transfer for subsequent partition archiving. The pending compilation buffer is scanned when the recycling trigger condition is met or the memory threshold is reached. The system archives the record entries in it according to the partition tag field into multiple parallel partition storage areas. Each partition storage area is a sequential file for a single partition index number. The system explicitly maintains the original time sequence when writing. The reason is that irrigation decisions must accurately reproduce the actual time sequence of node sampling and do not allow the causal chain of information to be disrupted due to disk write reordering. Subsequently, the central control terminal constructs row vectors in each partition storage area according to the preset field order. The column layout of the row vectors strictly follows the fixed template of five core environmental indicators plus the partition mark field and the sampling timestamp field. This template ensures that the same column always carries homogeneous information, which facilitates vectorized operations. After all row vectors are written to the preprocessing queue, the system performs an ascending order based on the sampling timestamp field at the partition level, so that the row vectors present an increasing time axis in the vertical dimension. The sorted row vectors are written into the matrix row cache in sequence, and the matrix row cache creates a partition row segment for each partition in the form of a block array.

[0042] After all partition rows are written, the system merges them continuously within the same memory page in ascending order of partition index numbers, ultimately generating a reorganized cotton field environmental information matrix. The row index set of this matrix consists of the Cartesian product of the partition index number and the sampling timestamp field, while the column index set encompasses all environmental indicator fields collected by the sensors. This ensures that the matrix possesses logically identical row and column isomorphism and isodimensionality, enabling it to be directly used as an input tensor by the vortex pulse weaving probability calculus model. Through this layered mapping from stream to table and from table to matrix, the central control terminal decomposes massive amounts of asynchronous data into a symbolic, serialized, and block-based multi-level structure. This ensures that each plot number corresponds to a unique partition index number at any given moment and that the data maintains monotonic continuity across both spatial and temporal coordinates. Ultimately, the real-world partition management semantics of the cotton field are embedded in the digital matrix, laying a highly consistent and traceable structural foundation for the subsequent construction of the cotton field collaborative data volume, topological mapping of the vortex hierarchy, and generation of irrigation time window lists and irrigation water volume lists.

[0043] Furthermore, to enable the model to uniformly map real-time environmental conditions and historical yield performance into the same data structure, the central control terminal needs to merge the reorganized cotton field environmental information matrix and the historical cotton field yield matrix into a collaborative cotton field data volume. The key to this fusion lies in establishing a strict correspondence between the temporal and spatial dimensions. First, the central control terminal adds a time-pairing field to each record in the historical cotton field yield matrix, uniformly converts the harvest year field to a timestamp for the end of that year, and writes this into the time-pairing field. This step converts the yield information, which was originally discrete by year, into the same timeline measurement system as the environmental data, avoiding empty or duplicate matches during subsequent joins due to inconsistent time granularity. The central control terminal then indexes the reorganized cotton field environmental information matrix in ascending order by the plot number field and the sampling timestamp field, and also indexes the historical cotton field yield matrix in ascending order by the plot number field and the time-pairing field.

[0044] The benefit of the double ascending index is that it ensures that the two matrices follow monotonicity in both spatial and temporal dimensions. During inner joins, the system only needs a linear scan to complete the match, without the need for additional hashing or tree structure acceleration, which saves memory and reduces latency. The inner join operation uses the plot number field and the time dimension field to form a composite key. The row vectors are merged only when the two keys are equal at the same time. During the merging process, the environmental indicator field maintains the high-frequency sampling value from the reorganization, and the yield indicator field maintains the annual yield value from the historical matrix. The newly generated row vector contains both short-term environmental change information and long-term yield reflection information. The cotton field collaborative data body obtained in this way maintains the same time order as the reorganized cotton field environmental information matrix in the row dimension, and expands the yield-related fields in the column dimension to form a unified tensor that can be directly input into the vortex pulse weaving probability calculation model. Through this generation process, the central control terminal integrates data of different time scales, different sampling frequencies, and different statistical meanings into a single coordinate system, enabling subsequent algorithms to simultaneously refer to the immediate status of partition moisture and years of historical yield feedback in a single tensor operation, and then output a dynamic water demand forecast sequence for cotton fields that is more in line with the actual water requirements of crops, providing a reliable data basis for the generation of irrigation time window lists and irrigation water volume lists.

[0045] Furthermore, step 3 specifically includes:

[0046] Step 3.1: Construct multiple vortex hierarchies and map the cotton field collaborative data volume into the inner vortex layer and the outer vortex layer in the spatial dimension and the temporal dimension respectively;

[0047] Step 3.2: Trigger a pulsed weaving operation to perform complementary winding of the node weights in the inner and outer vortex layers in an alternating order;

[0048] Step 3.3: After each weaving cycle, select the surviving weighted path based on the probability drift rule;

[0049] Step 3.4: Continue iterating until the preset convergence threshold is met to output the dynamic water demand prediction sequence of the cotton field.

[0050] The construction of a multi-vortex hierarchy is the starting point for the inference phase of the vortex pulse weaving probabilistic calculus model. Its principle can be likened to simultaneously unfolding a spatial grid and a temporal ribbon in a three-dimensional coordinate system. Each row vector in the cotton field collaborative data volume is then mapped to these two structures using a double-key mapping rule: the plot number field and the partition mark field are hashed to obtain the node coordinates of the inner vortex layer, while the sampling timestamp field is linearly indexed to obtain the node coordinates of the outer vortex layer, thus discretizing and anchoring the data in both spatial and temporal dimensions. The inner and outer vortex layers are essentially a pair of isomorphic sparse graphs, differing only in their projection axes. Together, they form the first level of the multi-vortex hierarchy, recursively expanding to deeper levels as the data volume increases. Each level has its own index number and maintains a one-to-one correspondence with the original row vectors. The mathematical motivation for this approach is to split the high-dimensional tensor into two orthogonal subspaces, allowing spatial coupling and temporal evolution to be topologically separated before reinteracting through subsequent weaving operations, thereby mitigating the negative impact of the curse of dimensionality on model weight convergence.

[0051] After hierarchical solidification is complete, the central control terminal injects an initial pulse signal. The pulse is expressed as discrete cycles in time and as a synchronized broadcast in space. It simultaneously activates all nodes in the inner and outer vortex layers and assigns an initial node weight to each node. The core idea of ​​complementary winding is to alternately allow inner and outer vortex nodes to serve as mutual weight gain sources. When the pulse count is in an odd cycle, the system traverses each node in the inner vortex layer and selects an outer vortex node with a mapping relationship based on the vortex connection table. The current weight of the other node is converted into an increment of its own weight using a complementary mapping function, thereby enhancing spatial information. When the pulse count is in an even cycle, the process is reversed, and the outer vortex node traverses and absorbs the weight of the inner vortex node, completing the enhancement of temporal information. The complementary mapping function is a linear normalization mapping under the logarithmic entropy constraint. It ensures that the sum of the weights of the two nodes remains conserved after a winding operation. The complementary ratio is determined by the local density of the nodes and the historical fluctuation amplitude. This allows high-confidence nodes to gradually dominate the weight distribution over multiple rounds of winding, while low-confidence nodes are probabilistically diluted. Whenever a pulse counting cycle ends, the central control terminal will sample the weight snapshot of all current nodes, extract the node weight drift path matrix sorted by spatial dimension index key and time dimension index key, and then select the surviving weight path according to the probability drift rule.

[0052] The probabilistic drift rule is designed as a dual-threshold gating mechanism: the system first calculates the fuzzy membership value of each path to measure its stable contribution to the overall output, and then calculates the path's cumulative drift probability to measure its information increment in the current iteration. These two values ​​are linearly coupled to produce a fuzzy entropy drift index. If this index exceeds an upper threshold, the path is immediately marked as a surviving weight path, meaning that this set of weight flows must be retained in subsequent iterations to maintain the information backbone. If the index falls between the two thresholds, a local time window stability test is performed. By comparing the short-term fluctuation variance with the long-term trend variance, the path is promoted to a candidate surviving weight path. Successful promotion is then treated as a surviving weight path in the next cycle. If the index falls below the lower threshold, the path is directly pruned, and the node weights are reset to zero and placed in a cooling queue to reduce the delay caused by noise on network convergence. This process repeats after each pulse cycle, forming an iterative survival competition until the global number of surviving weight paths remains unchanged for several consecutive cycles and the maximum increment of node weights falls below a preset convergence threshold. Only then does the system determine convergence success and extract the latest weight values ​​of all nodes in the surviving weight paths.

[0053] In order to convert these discrete node weights into a water demand curve on a continuous time axis, the central control terminal rearranges the node weights according to the spatial index and time index, and then calls the smoothing processing module to perform time series smoothing based on cubic spline interpolation, so that the interpolation curve passes through the original weight points of the high-confidence nodes and maintains the continuity of the second-order derivative to avoid overfitting oscillations. The interpolated curve constitutes a dynamic water demand prediction sequence for cotton fields, which covers the future schedulable interval in the time dimension and all partition index numbers in the spatial dimension. Each curve point identifies the estimated water demand for the corresponding partition at the corresponding moment. The sequence is then handed over to the resource allocation module to align with the available water resources to generate an irrigation time window list and an irrigation water volume list, thereby transforming the complex multi-dimensional vortex reasoning results into irrigation instructions that can be executed in the field. The essence of the entire step lies in organically connecting the three processes of cotton field collaborative data body splitting and mapping, pulse winding, and entropy-driven screening, so that spatial coupling and temporal evolution rotate and blend topologically, control the direction and intensity of information flow through probabilistic drift rules, and use iterative convergence thresholds to lock output stability at the macro level, ultimately providing cotton field moisture management with an intelligent decision-making basis that responds to the real-time environment and inherits historical laws.

[0054] Furthermore, the construction of multiple vortex levels elevates the cotton field collaborative data volume from a common two-dimensional table to a sparse graph topology capable of supporting the operation of the vortex pulse weaving probabilistic calculation model. The central control terminal first sequentially scans the cotton field collaborative data volume, simultaneously reading the plot number field, sampling timestamp field, and partition mark field for each row vector encountered. The plot number field and partition mark field semantically represent the agricultural operation unit and irrigation control unit, respectively. These two fields are concatenated to form a spatial dimension index key. This index key design ensures that the same plot maintains unique identity across different partitions, allowing multiple plots to form local clusters within the same partition, thereby providing hash continuity for subsequent spatial dimension mapping. The sampling timestamp field naturally has a globally monotonically increasing property and is therefore directly designated as the time dimension index key to maintain the physical order of row vectors along the time axis. To ensure symmetric memory mapping, the central control terminal creates a new multi-vortex hierarchical buffer during startup and pre-populates the hierarchical indexes of the first inner and outer vortex layers within this buffer, denoted as inner vortex index 1 and outer vortex index 1, respectively. Simultaneously, the system creates a hierarchical mapping table, using the hierarchical index as the primary key and the vortex depth as the value. This ensures that any node can be located at its corresponding level and depth in constant time through key-value reverse tracing, avoiding the stack overhead associated with recursive lookups.

[0055] When the central control terminal processes each row vector, it first calls the spatial dimension mapping logic, using the spatial dimension index key as input to generate the local number of the inner vortex node in this layer through a conflict-free hash function. This number is then combined with the current level index number to form a globally unique inner vortex node number, and the node is then written to the inner vortex layer node buffer. Next, the system executes the time dimension mapping logic on the same row vector, feeding the time dimension index key into the order mapping function to obtain the local number of the outer vortex node in this layer. This is also spliced ​​with the current level index number to form a globally unique outer vortex node number, and is then written to the outer vortex layer node buffer. Here, the two sets of mapping functions are independent of each other, allowing spatial information and temporal information to be expanded in parallel along their respective orthogonal dimensions, but both carry the original row number when written to facilitate the establishment of a connection later. The central control terminal then establishes a one-to-one node connection relationship between the newly generated inner vortex node number and the outer vortex node number based on the row number, and writes this correspondence into the vortex connection table in real time. The connection table stores tuples in the form of key-value pairs in memory. The key at one end points to the inner vortex node number, and the value at the other end points to the outer vortex node number, forming a lightweight pointer structure that can be traversed in both directions, providing a zero-copy index for the subsequent complementary winding stage.

[0056] As row vectors continue to be mapped, the number of nodes in the inner and outer vortex node buffers continues to grow. When the number of nodes on either side reaches a preset level expansion threshold, the central control terminal immediately creates the next inner and outer vortex layers in the multi-vortex level construction buffer, assigns them a new level index, and registers this index and the corresponding vortex level depth in the level mapping table. The level expansion threshold is determined based on the hardware memory capacity and the expected sparsity of the model, ensuring that the total number of nodes within a single layer is controlled, thereby keeping the hash table load factor within an acceptable range and avoiding query degradation. After the new layer is established, the system moves the level index pointer to the new layer and returns to the mapping logic to continue processing subsequent row vectors, recursively expanding the layers in batches until the entire cotton field collaborative data volume is mapped. There is no global lock contention during the entire mapping process because the write buffers for the spatial and temporal dimension mappings are independent. Pointer writes are performed only at the connection table level, and pointer writes are supported by atomic appends supported by lock-free linked lists, ensuring linear performance scalability under high concurrency.

[0057] When all row vectors have been mapped, the central control terminal starts an integrity check on the buffer zone of the multi-vortex hierarchy. The inspection process first counts the node counts of the inner vortex node buffer and the outer vortex node buffer, and checks whether they correspond one-to-one with the number of tuples in the vortex connection table; then scans the hierarchy mapping table to confirm that each hierarchy index number has at least one inner vortex node and one outer vortex node established in the buffer to prevent the occurrence of empty hierarchies; finally, a hash conflict detection is performed on the local numbering range of nodes in each layer to ensure that the mapping function does not generate key collisions in the current data set. After passing the integrity check, the system writes the three core data blocks of the inner vortex node buffer, the outer vortex node buffer and the vortex connection table to the disk in a sequential manner to the vortex hierarchy storage area, and records the hierarchy solidification timestamp at the same time. The hierarchy solidification timestamp is not only used to identify the model version, but also used as a boundary condition to distinguish between new and old node numbers in subsequent online incremental learning. At this point, the multi-vortex hierarchy construction process has been completed, and the cotton field collaborative data body has been successfully transformed into a sparse topological system consisting of several horizontal inner vortex layers, several vertical outer vortex layers, and cross-layer vortex connection tables, laying a rigorous and traceable data structure foundation for the space-time complementary winding iteration of the vortex pulse weaving probability calculation model.

[0058] The equation group for completing the numbering of inner and outer vortex nodes and automatically determining the vortex level of the i-th row vector of the cotton field collaborative data volume is: Among them, gid i The plot number field of the i-th row vector. i The partition mark field for the i-th row vector. i The sampling timestamp field of the i-th row vector. sp(·) is a collision-free spatial hash function, and the output range is [0,T ext -1]. rank τ (·) is the timestamp order mapping function, which converts the timestamp into a relative sequence number in the current layer, and the output value range is [0, T ext -1]. T ext is the layer expansion threshold, which indicates the maximum number of nodes that a single layer can accommodate. is the cumulative unique number of different spatial keys in the inner vortex layer node buffer when processing the i-th row. is the cumulative unique number of different time keys in the outer vortex node buffer when processing the i-th row. is the inner vortex level index number obtained by mapping the i-th row. is the outer vortex level index number obtained by mapping the i-th row. is the inner vortex node number corresponding to the i-th row, which is globally unique in hierarchical order. is the number of the outer vortex node corresponding to the i-th row, which is globally unique in hierarchical order. i It is the node connection tuple in the vortex connection table in row i, which contains the one-to-one corresponding inner vortex node number and outer vortex node number, and is used for fast indexing of subsequent weight winding and pulse weaving operations.

[0059] Furthermore, spatial dimension mapping and temporal dimension mapping jointly determine the node coordinate stability and search efficiency of subsequent pulse winding iterations. Spatial dimension mapping begins with a spatial dimension index key consisting of a plot number field and a partition tag field. This index key logically retains both field management unit information and the partition semantics of the irrigation valve control unit. Therefore, it must form an injective relationship with the inner vortex index. A hierarchical mapping table maintained by the central control terminal acts as a sparse index directory. Whenever the system intercepts a new spatial dimension index key, it queries the table in constant time. If no match is found, indicating that the spatial partition is not yet registered in the current hierarchy, the system assigns a new inner vortex index and writes the key-value pair back to the hierarchical mapping table. This write-back operation uses an atomic auto-incrementing pointer to prevent inner vortex index conflicts caused by concurrent allocations from multiple threads. The system then binds the spatial dimension index key of the row vector to the obtained inner vortex index and writes it, along with the original row number, to the inner vortex node buffer. This allows the unique spatial dimension index key to be traced at any time simply by the index number, ensuring spatial locality and enabling fast node location during search. The time dimension mapping is expanded based on the sampling timestamp field of the same row vector. The time dimension index key is naturally monotonically increasing. The central control terminal still queries the hierarchical mapping table first. If it is missing, a new outer vortex index number is assigned and registered. After registration, the time dimension index key and the outer vortex index number are bound and written into the outer vortex layer node buffer to form an outer vortex node. The same row vector thus occupies a unique node position in both the inner and outer vortex layers. The original row number is used as the natural primary key between the two nodes and is subsequently written directly into the vortex connection table, forming a one-to-one node connection relationship. Because the growth strategies of the inner and outer vortex index numbers are both controlled by the hierarchical expansion threshold, when the number of nodes on either side approaches the threshold, the system immediately switches to the next layer and refreshes the hierarchical pointer of the mapping table, achieving synchronous layering of the two orthogonal chains of spatial dimension and time dimension. This design allows new and old nodes to be distributed at different levels to avoid hash conflicts, while maintaining global connectivity through the connection table. This enables the vortex pulse weaving probability calculation model to accurately find paired nodes through single-hop indexing in subsequent complementary winding, fundamentally reducing the pressure of large-scale real-time irrigation data streams on memory fragmentation and retrieval delays, and ensuring that the entire method still maintains linear scalability under the high-frequency sampling environment of production fields.

[0060] Furthermore, the central control terminal first generates an initial pulse signal at the zero phase of the physical clock and simultaneously writes the constructed multiple vortex layers through broadcasting, so that all nodes in the inner and outer vortex layers enter the waiting state for weaving. The broadcast pulse is essentially a very short flag bit stream. Once perceived by the node, it immediately triggers a state transition. The node then adds a node weight field to its own storage area and writes the initial node weight. The initial node weight consists of two parts. One part comes from the linear combination of moisture-related indicators in the row vector of the cotton field collaborative data body represented by the node, and the other part is a global constant used to ensure that the new node has a non-zero weight before the first complementary winding. Next, the central control terminal starts to execute the complementary winding cycle according to the preset complementary winding order, with the inner vortex layer first, the outer vortex layer second, and the two alternating strategies. Each cycle first traverses the inner vortex layer. During the traversal, the nodes are accessed in reverse order according to their connectivity in the vortex connection table to increase the traction of high-connectivity nodes on the global weight distribution. An inner vortex node uses the connection table to find its corresponding outer vortex node, reads the outer vortex node's current weight, and then proportionally adds it to its own weight. This weighted average is then used to complete the complementary winding update. This update adheres to a conservation constraint, ensuring that the final weight of the inner vortex node is equal to the average of its own weight and the weight of the outer vortex node, proportioned according to connectivity. This allows information to be transferred from the time dimension to the spatial dimension without increasing global energy. After all nodes in the inner vortex layer have completed winding, control flow switches to the outer vortex layer. The outer vortex node follows the same process, reads the weight of the corresponding inner vortex node, and performs complementary winding, feeding spatial information back into the time dimension. These two stages constitute a complete complementary winding cycle. When the cycle counter increments a preset number of times, the node weights have probabilistically converged to a distribution characteristic of both spatial structure and time series. Through this cycle, the system ensures that each pulse drive is accompanied by a bidirectional exchange of weights, gradually reducing the influence of initial state randomness and strengthening the dominance of high-confidence nodes over the overall weight field. This ensures that the final node weights not only carry local information about the real-time environmental moisture deficit but also reflect the global trend of cumulative evolution over time.

[0061] The formula for synchronously updating the weights of the inner vortex node and the outer vortex node after a complementary winding operation is:

[0062]

[0063] in, is the node weight of the i-th node in the inner vortex layer at the beginning of the t-th winding cycle. is the node weight of the j-th node in the outer vortex layer at the beginning of the t-th winding cycle. is the new weight of the same inner vortex node after the complementary winding ends. is the new weight of the same outer vortex node after the complementary winding ends. is the connectivity of the i-th node in the inner vortex layer in the vortex connection table, that is, the number of outer vortex nodes that are connected with it. is the connectivity of the jth node in the outer vortex layer, that is, the number of inner vortex nodes that are connected to it. σ(i) is a mapping function that returns the outer vortex node index j that corresponds one-to-one to the inner vortex node i. -1 (j) is the reverse mapping function, which returns the inner vortex node index i corresponding to the outer vortex node j. t is the winding cycle counter, which is a non-negative integer and is used to identify the current iteration round.

[0064] Furthermore, at the end of the pulse weaving cycle, the central control terminal synchronously traverses the inner and outer vortex layers and writes the current node weights of all nodes into the node weight drift path matrix, according to the Cartesian order of their spatial and temporal index keys. Each row in this matrix represents a sequence of node weight paths that can be tracked across cycles. The system first calculates the path entropy vector for each path in the matrix. Path entropy measures the dispersion of node weight distribution within a path. Lower dispersion indicates a high concentration of information among a few key nodes, making it more valuable for reference. The path entropy vector is then fed into a fuzzy membership evaluation function. The output fuzzy membership value ranges from zero to one, with higher values ​​indicating a greater relative stability contribution of the path to the global output. In parallel, the central control terminal collects the weight change amplitude of each node weight path within the most recent pulse cycle. The cumulative amplitude of the node weight change is then accumulated and summed within a preset observation window. After normalization with the maximum cumulative amplitude among the same batch of paths, a drift probability value is obtained, ranging from zero to one. Higher values ​​indicate that the path carries more new information but also greater volatility. The system linearly couples the fuzzy membership value and the drift probability value to derive a fuzzy entropy drift index. This linear coupling maintains consistency between the two dimensions and highlights paths that meet the criteria of both stability and novelty. The fuzzy entropy drift index is then compared with preset upper and lower thresholds. Paths with an index greater than or equal to the upper threshold are directly entered into the surviving weighted path sequence, ensuring that key information is not eliminated. Paths with an index falling within the interval enter the secondary discrimination process, where the central control terminal performs stability checks on their fluctuation variance and trend variance within a local time window. If these tests pass, they are temporarily retained in the surviving weighted path sequence to observe their subsequent performance. Paths with an index less than the lower threshold are immediately deleted to reduce noise interference with model convergence. Through this combination of entropy measurement, fuzzy evaluation, and probability drift, the system adaptively slims down the path pool after each iteration, ensuring that important spatial-temporal weighted pathways continue to participate in the next round of complementary winding while gradually eliminating locally noisy paths.

[0065] Fuzzy entropy drift index E i for:

[0066]

[0067] Among them, E i is the fuzzy entropy drift index of the i-th node weight path, an intermediate variable. i is the path entropy of the i-th node weight path, an intermediate variable, and the calculation formula is where p i,k is the weight ratio of the path at the kth node. N i is the total number of nodes included in the i-th node weight path, an intermediate variable used to logarithmically normalize the path entropy. i (t) is the node weight of the i-th node weight path at the end of winding cycle t, an intermediate variable. i is the number of observation cycles when evaluating the drift probability of the i-th node weight path, an intermediate variable. is the weight path of the i-th node in T i The cumulative magnitude of weight changes within a period, intermediate variables. is the maximum value of the cumulative amplitude of weight changes in the same batch of paths, an intermediate variable, used to normalize the drift probability. This formula converts the fuzzy membership term (1-H i / logN i ) and the drift probability term obtained by the cumulative weight change Directly multiply to generate a single fuzzy entropy drift index E i , so that the central control terminal can compare with the upper and lower thresholds to complete the judgment of survival, pending survival or deletion.

[0068] Furthermore, the central control terminal first aggregates all node weights contained in the two types of path sequences in memory and rearranges them according to the Cartesian order of the spatial and temporal index keys, so that the weights corresponding to each timestamp within the same partition index number form a strictly monotonic time vector. The rearranged data remains discrete and exhibits short-period jitter, which, if directly used for irrigation scheduling, can result in frequent valve starts and stops. To address this, the system invokes a smoothing module to perform time series smoothing based on cubic spline interpolation on each rearranged time vector. Cubic spline interpolation constructs a B-spline basis function system at the original time nodes of each partition. The control node weight coefficients are obtained by solving a system of linear equations subject to natural endpoint constraints, generating a continuous curve with continuous second-order derivatives and gently varying curvature. This curve not only passes through all original sampling points, ensuring that model information is not biased, but also minimizes overall curvature to mitigate local peaks caused by isolated anomalous weights. After interpolation, the system samples the curve at a fixed time step to generate a series of dynamic water demand forecasts for the cotton field covering the future irrigation scheduling window.

[0069] The predicted water demand of the cotton field spatial unit with partition index number p at time point t, that is, a result in the cotton field dynamic water demand prediction sequence for:

[0070]

[0071] Among them, p is the partition index number in the spatial dimension index key, an intermediate variable used to distinguish different irrigation control units. t is the real time variable after the time dimension index key is continuous, an intermediate variable, and its value range is determined by the first and last two sampling timestamps t0 and t n-1 n is the total number of time nodes formed after the partition is rearranged in the survival weight path sequence and the waiting survival weight path sequence, an intermediate variable. p,i For partition p at time node t i The node weight after the above aggregation, the intermediate variable, is obtained by aggregating the weights of all corresponding nodes in the survival weight path sequence and the waiting survival weight path sequence according to the spatial dimension index key and the time dimension index key. i,3 (t) is the open node vector {t0,t0,t0,t0,t1,…,t n-1 ,t n-1 ,t n-1 ,t n-1} defined cubic B-spline basis function, the i-th basis function is in the interval [t i ,t i+4 ) takes a non-zero value, an intermediate variable, which is used to ensure that the entire interpolation curve is continuous in the second-order derivative.

[0072] The following is a complete example of generating a dynamic water demand prediction sequence for a cotton field using two plots and two time periods:

[0073] Plot number: gid1=G01, gid2=G02;

[0074] Partition mark: pid1 = P01, pid2 = P02;

[0075] Two sampling timestamps (milliseconds): t0 = 17582400000, t1 = 17582460000.

[0076] At time points t0 and t1, the following five indicators are received (unit order is volume fraction, degree Celsius, percentage, meter per second, watt per square meter):

[0077]

[0078] After the two nodes are synchronized to the same millisecond axis, no interpolation is required. The row order of the reorganized cotton field environment information matrix is ​​(s=0, t0)→(s=1, t0)→(s=0, t1)→(s=1, t1).

[0079] The yield (tons per hectare) of the previous quarter is Y(G01) = 6.30, Y(G02) = 6.05; the yield timestamp is unified to 16820160000 at the end of the previous year. After the inner join, each row vector has an additional column of fixed yield values. The hierarchical expansion threshold is set to T ext = 100. Calculate the first row vector (G01, P01, t0) For the second row vector The same logic is applied to the other two lines.

[0080] The initial weight formula is given by the linear combination of five indicators Taking (G01, P01, t0) as an example, M = 0.18, H = 0.72, L = 0.32, t = 29.5, R = 820; Same method

[0081] Inner vortex node and outer vortex node Complementary, the connectivity of both is d=1. Similarly, the weights of the other three pairs of nodes remain unchanged; if there is a difference in connectivity, the weights will tilt towards the node with higher connectivity.

[0082] The weights of the four paths remain unchanged after one winding, and the weight differences are all zero, so The fuzzy entropy drift index is lower than the lower threshold E low =0.10, but for demonstration purposes, all paths are retained in the waiting pool. A natural cubic spline is established for the two points (t0, 0.339) and (t1, 0.335) in the partition s=0. Since there are only two points, the curve degenerates into a linear Take the future moment Similarly, partition s=1 and we get

[0083] Assuming that the current available water resources are converted to the total demand weight of 0.550, then ΔQ0=0.331, ΔQ1=

[0084] 0.292,ΔQ Σ=0.623; the system detects an excess ratio of 0.623 / 0.550=1.133, and reduces the demand of the low priority partition s=1 by 0.073 to meet the resource constraint. Finally, the irrigation time window list (s=

[0085] 0,t2,0.331), (s=1,t2,0.219).

[0086] Figure 2 The detailed analysis results of the fuzzy entropy drift index and the survival weight path screening effect in the vortex pulse weaving probability calculation model are shown. The figure is in the form of a scatter distribution, with the horizontal axis representing the vortex level node number (number) and the vertical axis representing the fuzzy entropy drift index (dimensionless). Two key probability drift rule threshold lines are set in the figure: the upper threshold line is at 0.6 and the lower threshold line is at 0.4. These two threshold lines divide the entire fuzzy entropy drift index space into three different processing areas. In the area above the upper threshold line, there are 11 solid black dots, representing the survival weight path sequence. The fuzzy entropy drift index of these nodes is greater than or equal to the upper threshold of 0.6. According to the probability drift rule, these node weight path sequences are directly marked as survival states without further processing. From the distribution, it can be seen that the fuzzy entropy drift index of nodes 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, and 11 are 0.85, 0.75, 0.9, 0.95, 0.8, 0.92, 0.82, 0.78, 0.9, 0.75, and 0.85, respectively, which are significantly higher than the upper threshold standard.

[0087] In the area between the upper and lower thresholds, there are 10 hollow circles, representing weighted path sequences to be survived. The fuzzy entropy drift index of these nodes ranges from 0.4 to 0.6, requiring them to enter the secondary judgment process. Small squares in the figure indicate whether the local time window stability test has been passed. A total of six nodes passed the stability test and are marked as weighted path sequences to be survived. Specifically, the local time window stability test results for nodes 2, 3, 5, 6, 8, and 9 passed, and their corresponding weighted path sequences were retained; while the test results for nodes 1, 4, 7, and 10 failed, and their corresponding path sequences were deleted. In the area below the lower threshold, seven deleted path sequence points are marked with crosshairs. The fuzzy entropy drift index of these nodes is less than the lower threshold of 0.4, and the corresponding node weighted path sequences are directly deleted according to the probability drift rule. Experimental results show that this screening mechanism effectively optimizes weighted path selection, with surviving weighted path sequences accounting for 44% of the total, pending surviving weighted path sequences accounting for 24%, and deleted path sequences accounting for 32%. The overall screening efficiency meets the expected target.

[0088] Figure 3The accuracy improvement effect of the dynamic water demand prediction series of cotton fields before and after cubic spline interpolation smoothing is compared in detail. The figure uses a time series comparative analysis method, with the horizontal axis representing the sampling timestamp sequence (days) and the vertical axis representing the water demand (mm). The figure contains three sets of key data: actual cotton field water demand data points, the prediction series before smoothing, and the prediction series after cubic spline interpolation smoothing. The actual cotton field water demand data is marked with solid black dots and is distributed in the time series from the 1st to the 19th day. From the data distribution, it can be seen that the actual water demand shows obvious periodic fluctuation characteristics, reaching a minimum of about 45mm on the 7th day and a relative high of about 48mm on the 15th day. This fluctuation reflects the natural change law of water demand during cotton growth and is closely related to the physiological growth cycle of cotton.

[0089] The pre-smoothing forecast sequence, represented by the dotted line, is the original forecast generated directly from the survival weight path sequence, after reordering the spatial and temporal index keys. As can be seen from the figure, the forecast sequence exhibits significant volatility and discontinuity, resulting in significant deviations from the actual demand data. Error bar analysis reveals that the pre-smoothing forecast error is relatively large at all time points, with an average error of 8.7%. The maximum error occurs on the 15th day, with an error margin of 25 mm. The forecast sequence after cubic spline interpolation smoothing, represented by the thick solid line, exhibits good continuity and smoothness. The smoothing algorithm effectively eliminates abrupt changes and noise in the original forecast sequence, making the forecast curve more consistent with the actual changing trends in cotton field water demand. The smoothing results show a significantly improved fit between the smoothed forecast sequence and the actual demand data points, with significantly shorter error bars and a reduction in average forecast error to 2.3%, resulting in a 74% improvement in accuracy. Statistical analysis demonstrates that cubic spline interpolation smoothing not only improves forecast accuracy but also significantly reduces the convergence period by 60%. This improvement effect verifies the important role of the smoothing module in generating dynamic water demand prediction sequences for cotton fields, and provides a more reliable data basis for subsequent irrigation optimization decisions.

[0090] The present invention is not limited to the aforementioned specific embodiments, but extends to any new features or any new combination disclosed in this specification, as well as any new method or process steps or any new combination disclosed.

Claims

1. A cotton irrigation optimization method, characterized in that: The method comprises: Step 1: Through the multi-dimensional sensing nodes deployed in the target cotton field, soil moisture content, soil temperature, crop leaf surface humidity, wind speed and sunshine intensity are continuously acquired and uploaded to the central control terminal in real time to form the original cotton field environmental information set; Step 2: The central control terminal performs time-series synchronization processing on the original cotton field environmental information set, using timestamp alignment to eliminate cross-node sampling deviations to obtain time-series synchronized data; the time-series synchronized data is partitioned and marked based on the plot number to obtain the reorganized cotton field environmental information matrix; the historical cotton field yield matrix is ​​introduced into the reorganized cotton field environmental information matrix to generate a cotton field collaborative data volume; Step 3: Invoke the vortex pulse weaving probability calculation model in the central control terminal, perform decision calculations on the cotton field collaborative data volume, and output a dynamic water demand prediction sequence for the cotton field; Step 4: The central control terminal compares the current available water resources based on the dynamic water demand prediction sequence of the cotton field, uses the partition priority queue method to generate an irrigation time window list and a corresponding irrigation water volume list to complete the cotton irrigation optimization.

2. The cotton irrigation optimization method according to claim 1, wherein: In step 2, the process of partitioning and marking the time series synchronization data based on the plot number to obtain the reorganized cotton field environmental information matrix includes: the central control terminal reads the record entries one by one from the time series synchronization data, traverses the field set corresponding to each record entry, identifies and extracts the plot number field therein; the central control terminal generates a plot mapping table based on all the obtained plot number fields, which uses the plot number as the primary key and the sequentially increasing partition index number as the value, and performs a uniqueness check during the writing process to avoid the same plot number being repeatedly mapped to different partition index numbers; the central control terminal calls the plot mapping table and performs the following processing on each record entry in the time series synchronization data: reads the plot number field of the record entry; queries the plot mapping table to obtain the corresponding partition index number; a new partition mark field is created in the record entry, and the partition index number is written into it; the record entry after the partition mark field is written is stored in the buffer area to be reorganized; the central control terminal partitions and archives the record entries in the buffer area to be reorganized according to the partition mark field, and stores all record entries under the same partition mark field value in the same partition storage area in sequence; the original time sequence is kept unchanged during the storage process; the central control terminal constructs a row vector in each partition storage area according to the preset field order; all row vectors in the same partition storage area are arranged in ascending order according to the sampling timestamp field; the sorted row vectors in each partition storage area are written into the matrix row cache in sequence to form partition row segments; all partition row segments are merged in ascending order of the partition index number to generate the reorganized cotton field environmental information matrix.

3. The cotton irrigation optimization method according to claim 2, wherein: In step 2, the process of generating the cotton field collaborative data body includes: the central control terminal adds a time pairing field to each record in the historical cotton field yield matrix, converts the harvest year field into a unified timestamp of the last day of the year, and writes the time pairing field to form a time dimension correspondence with the sampling timestamp field in the reorganized cotton field environmental information matrix; the reorganized cotton field environmental information matrix is ​​indexed in ascending order according to the plot number field and the sampling timestamp field; the historical cotton field yield matrix is ​​indexed in ascending order according to the plot number field and the time pairing field, and an inner join operation is performed on the reorganized cotton field environmental information matrix and the historical cotton field yield matrix to obtain the cotton field collaborative data body.

4. The cotton irrigation optimization method according to claim 3, wherein: Step 3 specifically includes: Step 3.1: Construct multiple vortex hierarchies and map the cotton field collaborative data volume into the inner vortex layer and the outer vortex layer in the spatial dimension and the temporal dimension respectively; Step 3.2: Trigger a pulsed weaving operation to perform complementary winding of the node weights in the inner and outer vortex layers in an alternating order; Step 3.3: After each weaving cycle, select the surviving weighted path based on the probability drift rule; Step 3.4: Continue iterating until the preset convergence threshold is met to output the dynamic water demand prediction sequence of the cotton field.

5. The cotton irrigation optimization method according to claim 3, wherein: Step 3.1 specifically includes: the central control terminal reads the cotton field collaborative data body, and extracts the plot number field, sampling timestamp field and partition mark field for each row vector; wherein, the plot number field and partition mark field together constitute the spatial dimension index key, and the sampling timestamp field constitutes the time dimension index key; the central control terminal creates a new multiple vortex hierarchical structure buffer, and presets the hierarchical index numbers of the first inner vortex layer and the first outer vortex layer in the buffer, which are respectively recorded as inner vortex index number one and outer vortex index number one; at the same time, a hierarchical mapping table is created, with the hierarchical index number as the primary key and the vortex hierarchical depth number as the value; for each row vector in the cotton field collaborative data body, the central control terminal performs spatial dimension mapping to form an inner vortex node, and then performs time dimension mapping on the same row vector to form an outer vortex node; the buffer of the inner vortex layer node and the outer vortex layer node are written For the nodes in the node buffer, the central control terminal establishes a one-to-one corresponding node connection relationship based on the original row number of the row vector in the cotton field collaborative data body; the corresponding relationship is written into the vortex connection table; when the number of nodes in the inner vortex layer node buffer or the outer vortex layer node buffer reaches the preset level expansion threshold, the central control terminal creates the next inner vortex layer and the next outer vortex layer in the multiple vortex level construction buffer, and assigns a new level index number to it; then returns to execute spatial dimension mapping and time dimension mapping until the cotton field collaborative data body is fully mapped; the central control terminal performs an integrity check on the multiple vortex level construction buffer, and after confirming that the inner vortex layer node buffer, the outer vortex layer node buffer and the vortex connection table are fully recorded, all data is written into the vortex level storage area, and the level solidification timestamp is recorded to complete the construction of the multiple vortex level.

6. The cotton irrigation optimization method according to claim 5, characterized in that: The process of the central control terminal performing spatial dimension mapping specifically includes: for each row vector in the cotton field collaborative data body, the central control terminal performs spatial dimension mapping according to the spatial dimension index key, specifically including: if the corresponding inner vortex index number does not exist in the hierarchical mapping table, the central control terminal assigns a new inner vortex index number to the spatial dimension index key, and registers it in the hierarchical mapping table; the spatial dimension index key of the row vector is bound to the corresponding inner vortex index number and written into the inner vortex layer node buffer to form an inner vortex node; the process of the central control terminal performing time dimension mapping specifically includes: the central control terminal performs time dimension mapping on the time dimension index key of the same row vector, specifically including: if the corresponding outer vortex index number does not exist in the hierarchical mapping table, the central control terminal assigns a new outer vortex index number to the time dimension index key, and registers it in the hierarchical mapping table; the time dimension index key of the row vector is bound to the corresponding outer vortex index number and written into the outer vortex layer node buffer to form an outer vortex node.

7. The cotton irrigation optimization method according to claim 6, characterized in that: Step 3.2 specifically includes: the central control terminal generates an initial pulse signal, and simultaneously sends an initial pulse signal to the inner vortex layer and the outer vortex layer in the constructed multiple vortex levels, activates all nodes in the two vortex layers, and enters the waiting-for-weaving state; the central control terminal adds a node weight field to the nodes in each vortex layer, and assigns an initial node weight to each node; the central control terminal starts from the inner vortex layer according to the preset complementary winding order, and repeats the following operations alternately until a preset number of complementary winding cycles are completed: each node in the inner vortex layer selects a node in the corresponding outer vortex layer that has established a connection relationship, reads the node weight of the other party, and performs a complementary winding operation with its own node weight, and updates its own node weight; each node in the outer vortex layer selects a node in the corresponding inner vortex layer that has established a connection relationship, reads the node weight of the other party, and performs a complementary winding operation with its own node weight, and updates its own node weight.

8. The cotton irrigation optimization method according to claim 7, wherein: Step 3.3 specifically includes: after each weaving cycle, the central control terminal extracts the node weights of all nodes in the inner vortex layer and the outer vortex layer in turn, and arranges them in sequence according to the spatial and temporal indexes corresponding to the nodes to form a node weight drift path matrix; the central control terminal inputs the path entropy vector into the fuzzy membership evaluation function to obtain the fuzzy membership value; at the same time, the drift probability value is calculated based on the cumulative amplitude of the node weight change; the fuzzy membership value and the drift probability value are linearly coupled to obtain the fuzzy entropy drift index; the central control terminal presets two-level probability drift rule thresholds of upper threshold and lower threshold: when the fuzzy entropy drift index is greater than or equal to the upper threshold, the corresponding node weight path sequence is directly marked as a surviving weight path sequence; when the fuzzy entropy drift index is between the upper threshold and the lower threshold, the secondary judgment process is entered, and the local time window stability test is called. If the test passes, it is marked as a weight path sequence to be survived; when the fuzzy entropy drift index is less than the lower threshold, the corresponding node weight path sequence is directly deleted.

9. The cotton irrigation optimization method according to claim 8, wherein: Step 3.4 specifically includes: the central control terminal extracts all node weights in the survival weight path sequence and the to-be-survived weight path sequence, and rearranges the node weights according to the spatial dimension index key and the time dimension index key; then calls the smoothing processing module to perform time series smoothing based on cubic spline interpolation to complete the generation of the cotton field dynamic water demand prediction sequence.

10. A system for implementing the cotton irrigation optimization method according to any one of claims 1 to 9, characterized in that: The system includes: multidimensional sensing nodes and a central control terminal; the multidimensional sensing nodes are deployed in the target cotton field, continuously acquiring soil moisture content, soil temperature, crop leaf surface humidity, wind speed and sunshine intensity, and uploading them to the central control terminal in real time to form an original cotton field environmental information set; the central control terminal performs time series synchronization processing on the original cotton field environmental information set, adopts a timestamp alignment method to eliminate cross-node sampling deviation, and obtains time series synchronization data; the time series synchronization data is partitioned and marked based on the plot number to obtain a reorganized cotton field environmental information matrix; the historical cotton field yield matrix is ​​introduced into the reorganized cotton field environmental information matrix to generate a cotton field collaborative data body; the vortex pulse weaving probability calculation model is called to perform decision calculation on the cotton field collaborative data body and output a cotton field dynamic water demand prediction sequence; based on the cotton field dynamic water demand prediction sequence, the current available water resources are compared, and a partition priority queue method is adopted to generate an irrigation time window list and a corresponding irrigation water volume list to complete cotton irrigation optimization.

Images