Agricultural field water-salt evolution rule mining system based on time sequence trajectory clustering

Abstract

The present application relates to the field of agricultural data analysis, and more particularly to a farmland water-salt evolution law mining system based on time series trajectory clustering, which comprises the following steps: collecting and preprocessing original farmland water and salt data to obtain a two-dimensional time series trajectory set; analyzing the water-salt lag accumulation effect of the two-dimensional time series trajectory set to obtain a first optimization factor; analyzing the local water-salt trajectory fluctuation degree of the two-dimensional time series trajectory set to obtain a second optimization factor; jointly modifying the original single-step distance based on the first optimization factor and the second optimization factor to obtain a final time series similarity distance; and clustering and dividing the final time series similarity distance to extract the farmland water-salt evolution law, thereby solving the problem that the existing dynamic time warping algorithm cannot represent the asynchronous lag difference of farmland water and salt, resulting in the wrong clustering of different water-salt evolution mechanism plots.

Description

Technical Field

[0001] This invention relates to the field of agricultural data analysis technology, and in particular to a system for mining the evolution patterns of farmland water and salt based on time-series trajectory clustering. Background Technology

[0002] In the process of farmland water and salt regulation and saline-alkali land management, the temporal changes in soil moisture and salinity directly reflect the water and salt migration patterns under the influence of irrigation, rainfall, evaporation, and infiltration. To achieve refined management of large-scale farmland plots, it is typically necessary to continuously monitor soil moisture and salinity changes in different plots within a complete irrigation and evaporation cycle. Based on the monitoring results, the water and salt evolution types of different plots can be identified, thus providing data support for irrigation regulation, salt leaching and drainage, soil improvement, and zoned management. In existing technologies, the mining of water and salt evolution patterns across multiple farmland plots usually employs a combination of temporal trajectory analysis and cluster identification. This involves using the time series of moisture and salinity data for each plot as the analysis object, calculating the similarity between plots using temporal similarity metrics, and then combining this with clustering algorithms to classify the plots.

[0003] Among existing methods for calculating temporal similarity, the Dynamic Time Warping (VTW) algorithm, due to its ability to nonlinearly stretch and compress the time axis, can overcome, to some extent, the problem of overall time offset between different plots in terms of irrigation start time, rainfall response time, or evaporation process. Therefore, it is widely used in the alignment analysis of two-dimensional time-series water and salt trajectories in farmland. The VTW algorithm typically calculates the local distance between different time points and constructs a cumulative distance matrix. It then searches for the globally optimal alignment path among all possible matching paths to obtain the distance value between two time-series trajectories, which is then used as the basis for subsequent clustering. Based on this method, it is possible to compare and classify the water and salt evolution processes between different monitoring plots to a certain extent.

[0004] However, in real-world farmland scenarios, changes in soil moisture and salinity do not occur synchronously, but rather exhibit a significant asynchronous lag. Typically, irrigation or rainfall initially increases soil moisture, while salinity changes only gradually manifest after water infiltration, leaching, and redistribution. Different plots exhibit varying lag periods and evolution intensities in their water-salt responses due to differences in soil texture, pore structure, topsoil condition, and initial salinity distribution. Existing dynamic time warping algorithms, when calculating local distances, primarily align based on instantaneous numerical differences between time points, failing to reflect the lag relationship between water-driven salinity changes within different plots, and also struggling to characterize the differences in intervention intensity corresponding to varying degrees of local trajectory fluctuations. Consequently, when matching two-dimensional time-series water-salt trajectories across different plots, there is a tendency to overstretch or compress the time axis in pursuit of minimizing numerical errors. This can lead to plots with significantly different water-salt evolution mechanisms being incorrectly classified as similar, resulting in subsequent clustering results that are inconsistent with the actual water-salt migration patterns in farmland. Therefore, how to effectively characterize the lag differences and fluctuation differences in farmland water and salt evolution during the calculation of time-series trajectory similarity, thereby improving the accuracy of the results of farmland water and salt evolution law mining, has become an urgent technical problem to be solved. Summary of the Invention

[0005] In view of this, the present invention aims to propose a system for mining the evolution patterns of farmland water and salt based on time-series trajectory clustering, in order to solve the problem that existing dynamic time warping algorithms cannot characterize the asynchronous hysteresis differences of farmland water and salt, resulting in the incorrect clustering of plots with different water and salt evolution mechanisms.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A system for mining farmland water-salinity evolution patterns based on temporal trajectory clustering, the method comprising:

[0008] Step S1: Obtain a two-dimensional time-series trajectory set by collecting and preprocessing raw data on farmland moisture and salinity;

[0009] Step S2: Obtain the first optimization factor by performing water-salt hysteresis cumulative effect analysis on the two-dimensional time-series trajectory set;

[0010] Step S3: Obtain the second optimization factor by performing local water-salt trajectory fluctuation analysis on the two-dimensional time-series trajectory set;

[0011] Step S4: Obtain the final temporal similarity distance by jointly correcting the original single-step distance based on the first optimization factor and the second optimization factor;

[0012] Step S5: Obtain the evolution pattern of farmland water and salt by clustering and extracting patterns from the final temporal similarity distance.

[0013] Furthermore, the process of acquiring and preprocessing raw data on farmland moisture and salinity to obtain a two-dimensional time-series trajectory set includes:

[0014] A soil sensing sensor network is deployed in the target farmland area according to a preset spatial grid. Soil moisture sensor nodes for collecting raw soil moisture data and soil conductivity sensor nodes for collecting raw soil salinity data are set at the monitoring plots corresponding to each spatial grid. A fixed sampling frequency is set, and each monitoring plot is continuously monitored and collected according to the fixed sampling frequency. The collection time covers a complete irrigation and evaporation cycle to obtain the raw soil moisture data and raw soil conductivity data corresponding to each monitoring plot.

[0015] Based on the preset empirical conversion relationship between soil electrical conductivity and salinity, the raw soil electrical conductivity data of each monitoring plot are processed by salinity conversion to obtain the raw soil salinity data of each monitoring plot; wherein, the raw soil moisture data is the soil volumetric water content time series data, and the raw soil salinity data is the soil total salinity time series data.

[0016] Outlier removal, missing value completion, and time axis alignment were performed on the raw soil moisture and soil salinity data of each monitoring plot. The processed raw soil moisture and soil salinity data were then fused to obtain a set of two-dimensional time-series trajectories for each monitoring plot.

[0017] Furthermore, the step of obtaining the first optimization factor by performing water-salt hysteresis cumulative effect analysis on the two-dimensional time-series trajectory set includes:

[0018] By correlating and accumulating the moisture change data and salinity change data in the two-dimensional time-series trajectory set, the lagged cumulative amount of water and salinity is obtained.

[0019] The first optimization factor is obtained by applying a differential penalty to the lag accumulation of water and salt.

[0020] Furthermore, the step of obtaining the water-salinity lag accumulation by correlating and accumulating the moisture change data and salinity change data in the two-dimensional time-series trajectory set includes:

[0021] For any first target two-dimensional time-series trajectory to be clustered, the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory are extracted from the set of two-dimensional time-series trajectories. Then, the first-order difference calculation of adjacent time points is performed on the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory to obtain the moisture change data and salinity change data corresponding to the first target two-dimensional time-series trajectory.

[0022] For any first historical time node and any first current time node in the two-dimensional time-series trajectory of the first target, when the first current time node is not earlier than the first historical time node, the moisture change data corresponding to the first historical time node is used as the first moisture-driven assessment; for each time node between the first historical time node and the first current time node, the salinity change data corresponding to each time node is used as the corresponding salinity response assessment.

[0023] For any target time interval between the first historical time node and the first current time node, the square root of the sum of the squares of the moisture change data and the squares of the salinity change data at adjacent moments within the target time interval is taken, and the square root results corresponding to all adjacent moments are accumulated. The accumulated result is taken as the two-dimensional state space path length between the first historical time node and the first current time node. The salinity response assessment corresponding to each time node is divided by the sum of a constant and the two-dimensional state space path length, and the calculated results are accumulated to obtain the cumulative salinity decay response assessment. The result of multiplying the first moisture driving assessment and the cumulative salinity decay response assessment is taken as the hysteresis driving contribution corresponding to the first historical time node.

[0024] For the first current time node, the lag driving contribution of all the first historical time nodes before the first current time node is accumulated to obtain the water and salt lag accumulation of the first target two-dimensional time series trajectory at the first current time node.

[0025] For any second target two-dimensional time-series trajectory used for comparison and matching, the same processing method as the first target two-dimensional time-series trajectory is adopted to obtain the water and salt lag accumulation corresponding to any second current time node of the second target two-dimensional time-series trajectory.

[0026] Furthermore, the step of obtaining the first optimization factor by applying a difference penalty to the accumulated water-salt lag includes:

[0027] For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, obtain the water and salt lag accumulation corresponding to the first target two-dimensional time trajectory at the first current time node and the water and salt lag accumulation corresponding to the second target two-dimensional time trajectory at the second current time node.

[0028] The absolute value of the difference between the water and salt lag accumulation of the first target two-dimensional time series trajectory at the first current time node and the water and salt lag accumulation of the second target two-dimensional time series trajectory at the second current time node is used as the water and salt lag accumulation difference assessment.

[0029] The absolute value of the water and salt lag accumulation of the first target two-dimensional time trajectory at the first current time node is added to the absolute value of the water and salt lag accumulation of the second target two-dimensional time trajectory at the second current time node, and the addition result is added to a preset minimum positive constant. The corresponding calculation result is used as the normalization benchmark for the water and salt lag accumulation.

[0030] The difference assessment of water and salt lag accumulation is used as the numerator, the normalized benchmark of water and salt lag accumulation is used as the denominator, and the corresponding fraction is used as the relative difference assessment of water and salt lag accumulation.

[0031] The relative difference assessment of water and salt lag accumulation is processed by exponential mapping with the natural constant as the base, and the first optimization factor corresponding to the first target two-dimensional time series trajectory at the first current time node and the second target two-dimensional time series trajectory at the second current time node is obtained.

[0032] Furthermore, the step of obtaining the second optimization factor by performing local water-salt trajectory fluctuation analysis on the two-dimensional time-series trajectory set includes:

[0033] By extracting local fluctuation features from the moisture and salinity data in the two-dimensional time-series trajectory set, local water and salinity trajectory fluctuation feature data can be obtained.

[0034] The local water-salt trajectory fluctuation characteristic data is normalized and processed to obtain the local water-salt trajectory fluctuation degree.

[0035] A second optimization factor is obtained by applying a differential penalty to the local water-salt trajectory fluctuation.

[0036] Furthermore, the step of extracting local fluctuation features from the moisture and salinity data in the two-dimensional time-series trajectory set to obtain local water-salt trajectory fluctuation feature data includes:

[0037] For any first target two-dimensional time-series trajectory to be clustered, extract the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory from the set of two-dimensional time-series trajectories;

[0038] For any first current time node in the two-dimensional time-series trajectory of the first target, when the first current time node has the original soil moisture data and original soil salinity data corresponding to the previous time node and the two previous time nodes, the original soil moisture data corresponding to the first current time node is combined with twice the original soil moisture data corresponding to the previous time node and the original soil moisture data corresponding to the two previous time nodes to obtain the second-order moisture fluctuation assessment corresponding to the first current time node; the original soil salinity data corresponding to the first current time node is combined with twice the original soil salinity data corresponding to the previous time node and the original soil salinity data corresponding to the two previous time nodes to obtain the second-order salinity fluctuation assessment corresponding to the first current time node.

[0039] The square root of the sum of the square of the second-order fluctuation assessment of moisture and the square of the second-order fluctuation assessment of salinity at the first current time node is used to obtain the local fluctuation intensity assessment at the first current time node.

[0040] For any first current time node in the two-dimensional time-series trajectory of the first target, all the acquired local fluctuation intensity assessments are accumulated to obtain the local water and salt trajectory fluctuation characteristic data corresponding to the first current time node of the two-dimensional time-series trajectory of the first target.

[0041] For any second target two-dimensional time-series trajectory used for comparison and matching, the same processing method as the first target two-dimensional time-series trajectory is adopted to obtain the local water and salt trajectory fluctuation feature data corresponding to any second current time node of the second target two-dimensional time-series trajectory.

[0042] Furthermore, the step of obtaining the local water-salt trajectory fluctuation degree by normalizing the local water-salt trajectory fluctuation characteristic data includes:

[0043] For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, obtain the local water and salt trajectory fluctuation feature data corresponding to the first target two-dimensional time trajectory at the first current time node and the local water and salt trajectory fluctuation feature data corresponding to the second target two-dimensional time trajectory at the second current time node;

[0044] For any first current time node in the first target two-dimensional time-series trajectory, extract the original soil moisture data and original soil salinity data corresponding to the first current time node from the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory, and extract the original soil moisture data and original soil salinity data corresponding to the starting time node; add the square of the difference between the original soil moisture data corresponding to the first current time node and the original soil moisture data corresponding to the starting time node, and take the square root of the sum of the square of the difference between the original soil salinity data corresponding to the first current time node and the original soil salinity data corresponding to the starting time node to obtain the net displacement assessment of the first target two-dimensional time-series trajectory corresponding to the first current time node;

[0045] Using the local water and salt trajectory fluctuation characteristic data of the first target two-dimensional time-series trajectory at the first current time node as the numerator and the net displacement assessment of the first target two-dimensional time-series trajectory at the first current time node as the denominator, the local water and salt trajectory fluctuation degree of the first target two-dimensional time-series trajectory at the first current time node is obtained.

[0046] For any second current time node in the second target two-dimensional time-series trajectory, the same processing method as for the first target two-dimensional time-series trajectory is adopted to obtain the net displacement assessment of the second target two-dimensional time-series trajectory at the second current time node. The local water and salt trajectory fluctuation characteristic data of the second target two-dimensional time-series trajectory at the second current time node is used as the numerator, and the net displacement assessment of the second target two-dimensional time-series trajectory at the second current time node is used as the denominator to obtain the local water and salt trajectory fluctuation degree of the second target two-dimensional time-series trajectory at the second current time node.

[0047] Furthermore, the step of obtaining the second optimization factor by applying a difference penalty to the local water-salt trajectory fluctuation includes:

[0048] For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, obtain the local water and salt trajectory fluctuation of the first target two-dimensional time trajectory at the first current time node and the local water and salt trajectory fluctuation of the second target two-dimensional time trajectory at the second current time node.

[0049] The absolute value of the difference between the local water and salt trajectory fluctuation of the first target two-dimensional time-series trajectory at the first current time node and the local water and salt trajectory fluctuation of the second target two-dimensional time-series trajectory at the second current time node is used as the local water and salt trajectory fluctuation difference assessment.

[0050] The larger of the local water and salt trajectory fluctuation of the first target two-dimensional time-series trajectory at the first current time node and the local water and salt trajectory fluctuation of the second target two-dimensional time-series trajectory at the second current time node is added to a preset minimum positive constant, and the corresponding calculation result is used as the normalization benchmark for local water and salt trajectory fluctuation.

[0051] The local water-salt trajectory fluctuation difference assessment is used as the numerator, the local water-salt trajectory fluctuation normalization benchmark is used as the denominator, and the corresponding fraction is used as the local water-salt trajectory fluctuation relative difference assessment.

[0052] The relative difference assessment of local water-salt trajectory fluctuation is added to a constant 1 to obtain the second optimization factor corresponding to the first target two-dimensional time-series trajectory at the first current time node and the second target two-dimensional time-series trajectory at the second current time node.

[0053] Furthermore, the step of obtaining the final temporal similarity distance by jointly correcting the original single-step distance based on the first optimization factor and the second optimization factor includes:

[0054] For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, extract the original soil moisture data and original soil salinity data corresponding to the first current time node from the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time trajectory, and extract the original soil moisture data and original soil salinity data corresponding to the second current time node from the original soil moisture data and original soil salinity data corresponding to the second target two-dimensional time trajectory.

[0055] The square root of the square of the difference between the original soil moisture data corresponding to the first current time node and the original soil moisture data corresponding to the second current time node is obtained by adding the square of the difference between the original soil salinity data corresponding to the first current time node and the original soil salinity data corresponding to the second current time node.

[0056] The product of the original single-step distance and the first optimization factor and the second optimization factor is used as the corrected single-step distance between the first target two-dimensional time-series trajectory at the first current time node and the second target two-dimensional time-series trajectory at the second current time node.

[0057] For any first current time node in the two-dimensional time-series trajectory of the first target and any second current time node in the two-dimensional time-series trajectory of the second target, the smaller of the following values ​​is added: the correction step distance, the minimum cumulative distance corresponding to the previous time node of the first current time node and the second current time node, the minimum cumulative distance corresponding to the previous time node of the first current time node and the second current time node, and the minimum cumulative distance corresponding to the previous time node of the first current time node and the previous time node of the second current time node. This yields the minimum cumulative distance between the two-dimensional time-series trajectory of the first target at the first current time node and the two-dimensional time-series trajectory of the second target at the second current time node.

[0058] The minimum cumulative distance between the end time node of the first target's two-dimensional time-series trajectory and the end time node of the second target's two-dimensional time-series trajectory is taken as the final temporal similarity distance between the two-dimensional time-series trajectories of the first and second targets.

[0059] Compared with the prior art, the present invention has the following advantages:

[0060] The farmland water and salt evolution law mining system based on temporal trajectory clustering described in this invention introduces a constraint representation of the lag characteristics of water-driven salt change during the similarity measurement process of two-dimensional temporal trajectories of farmland water and salt. This allows for accurate differentiation of plots with similar surface numerical changes but different underlying evolutionary mechanisms, even in complex farmland scenarios with varying irrigation times, soil textures, and water and salt transport rates. Compared to existing methods that rely solely on instantaneous numerical deviations for time alignment, this technique effectively suppresses erroneous matching caused by excessive stretching or compression of the time axis. This prevents plots with short evolution cycles and rapid water and salt replacement from being misclassified as the same type as plots with long evolution cycles and slow water and salt replacement, significantly improving the consistency between the farmland water and salt evolution law identification results and the actual soil water and salt migration process. This provides a more reliable basis for irrigation regulation, salt drainage management, and zoning management. Furthermore, this invention combines the differences in the intensity of local fluctuations in two-dimensional trajectories to precisely identify the differences in the kinetic energy of water-salt evolution under different intervention conditions. This effectively avoids smoothing out and conflating the drastic water-salt fluctuations generated under strong irrigation conditions with the gentle seepage processes formed under weak intervention conditions. Using this technology, the system can not only identify whether plots have similar water-salt response stages, but also further identify the intensity and dynamic differences in their evolution processes. Therefore, the output clustering results have stronger physical interpretability and engineering applicability. For large-scale farmland monitoring and precision management, this technical solution is conducive to generating more realistic and detailed classification results of water-salt evolution types, facilitating managers to formulate differentiated irrigation, salt leaching, and soil improvement strategies for different types of plots. Attached Figure Description

[0061] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0062] Figure 1 This is a system flowchart of the system for mining the evolution law of farmland water and salt based on time-series trajectory clustering, as described in an embodiment of the present invention. Detailed Implementation

[0063] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0064] See Figure 1 This is a system flowchart of the farmland water and salt evolution law mining system based on time-series trajectory clustering provided in Embodiment 1 of the present invention, as follows: Figure 1 As shown, a system for mining farmland water-salinity evolution patterns based on time-series trajectory clustering may include:

[0065] Step S1: Obtain a two-dimensional time-series trajectory set by collecting and preprocessing the raw data of farmland moisture and salinity.

[0066] A soil sensing sensor network is deployed according to a preset spatial grid within the target farmland area. Soil moisture sensor nodes for collecting raw soil moisture data and soil conductivity sensor nodes for collecting raw soil salinity data are set at the monitoring plots corresponding to each spatial grid. In this embodiment of the invention, the size of the spatial grid is set to 100 meters long and 100 meters wide. A fixed sampling frequency is set, preferably once per hour. Continuous monitoring and data collection are performed on each monitoring plot according to the fixed sampling frequency, covering a complete irrigation and evaporation cycle, to obtain raw soil moisture and soil conductivity data for each monitoring plot.

[0067] Based on a pre-defined empirical conversion relationship between soil electrical conductivity and salinity, the raw soil electrical conductivity data for each monitoring plot were processed to convert salinity, thus obtaining the raw soil salinity data for each plot. The raw soil moisture data were time-series data of soil volumetric water content, and the raw soil salinity data were time-series data of total soil salinity.

[0068] It should be noted that the embodiments of the present invention use existing conversion formulas for soil electrical conductivity and salinity content. Specifically, ,in Indicates the first Soil total salinity data calculated at each time point Indicates the first Soil electrical conductivity data collected at each time point. Indicates the salt conversion coefficient of the regional soil. The salt transformation intercept constant of the regional soil is set in the embodiments of the present invention. ; .

[0069] Outlier removal, missing value completion, and time axis alignment were performed on the raw soil moisture and soil salinity data of each monitoring plot. The processed raw soil moisture and soil salinity data were then fused to obtain a set of two-dimensional time-series trajectories for each monitoring plot.

[0070] It should be noted that, for outlier removal of the original data, this embodiment of the invention uses the three-standard-deviation principle for outlier removal. For missing value imputation of the original data, this embodiment of the invention uses linear interpolation for missing value imputation. For time axis alignment, this embodiment of the invention uses a unified start time as the basis for time axis alignment.

[0071] Thus, the acquisition of a two-dimensional time-series trajectory set by collecting and preprocessing raw data on farmland moisture and salinity has been completed.

[0072] Step S2: Obtain the first optimization factor by performing a water-salt hysteresis cumulative effect analysis on the two-dimensional time-series trajectory set.

[0073] In real-world farmland scenarios, a dramatic increase in soil moisture (irrigation) drives a decrease in salinity (leaching). However, this driving force is not instantaneous but exhibits a significant time lag. More importantly, due to differences in soil porosity and texture, the effectiveness of water infiltration in hindering salt transport varies drastically across different plots, directly resulting in varying lengths of water-salt hysteresis cycles in different sequences. When traditional dynamic time warping algorithms process this type of data, to minimize the overall distance between the two time series, the algorithm alters the time step, forcibly aligning the rapidly occurring narrow water peaks and salinity troughs in fast-infiltration plots with the slowly occurring wide waveforms in slow-infiltration plots. This pathological timeline stretching, purely for numerical fitting, masks the inherent water-salt coupling hysteresis characteristics between plots. To prevent this erroneous alignment, this step addresses this by checking whether the two trajectory points to be aligned are in similar hysteresis response stages during their respective evolutionary cycles when calculating the sequence matching path. We need to abandon isolated instantaneous point comparisons and instead quantify the cumulative driving effect of previous moisture fluctuations on subsequent salinity changes at the current moment. Alignment should only be allowed when the cumulative driving effects behind two trajectory points are similar; otherwise, a significant alignment cost penalty should be imposed.

[0074] In summary, this invention first obtains the water-salt lag accumulation by correlating and accumulating moisture and salinity change data in a two-dimensional time-series trajectory set. Specifically, for any first target two-dimensional time-series trajectory to be clustered, the original soil moisture and soil salinity data corresponding to the first target two-dimensional time-series trajectory are extracted from the two-dimensional time-series trajectory set. Then, first-order difference calculations are performed on the original soil moisture and soil salinity data corresponding to the first target two-dimensional time-series trajectory at adjacent time points to obtain the moisture and salinity change data corresponding to the first target two-dimensional time-series trajectory. For any first historical time node and any first current time node in the first target two-dimensional time-series trajectory, when the first current time node is not earlier than the first historical time node, the moisture change data corresponding to the first historical time node is used as the first moisture-driven evaluation. For each time node between the first historical time node and the first current time node, the salinity change data corresponding to each time node is used as the corresponding salinity response evaluation. For any target time interval between the first historical time node and the first current time node, the square root of the sum of the squares of the moisture change data and the squares of the salinity change data at adjacent moments within the target time interval is taken, and the square root results for all adjacent moments are accumulated. The accumulated result is used as the two-dimensional state space path length between the first historical time node and the first current time node. The salinity response assessment corresponding to each time node is divided by the sum of a constant and the two-dimensional state space path length, and the calculated results are accumulated to obtain the salinity decay cumulative response assessment. The result of multiplying the first moisture-driven assessment and the salinity decay cumulative response assessment is used as the lag-driven contribution corresponding to the first historical time node. For the first current time node, the lag-driven contributions corresponding to all first historical time nodes before the first current time node are accumulated to obtain the water and salinity lag accumulation of the first target two-dimensional time series trajectory at the first current time node. For any second target two-dimensional time series trajectory used for comparison and matching, the same processing method as for the first target two-dimensional time series trajectory is used to obtain the water and salinity lag accumulation of the second target two-dimensional time series trajectory at any second current time node.

[0075] After obtaining the water and salt lag accumulation, a difference penalty is applied to the water and salt lag accumulation to obtain the first optimization factor. Specifically, for any first current time node in any first target two-dimensional time series trajectory to be aligned and any second current time node in any second target two-dimensional time series trajectory, the water and salt lag accumulation corresponding to the first target two-dimensional time series trajectory at the first current time node and the water and salt lag accumulation corresponding to the second target two-dimensional time series trajectory at the second current time node are obtained. The absolute value of the difference between the water and salt lag accumulation corresponding to the first target two-dimensional time series trajectory at the first current time node and the water and salt lag accumulation corresponding to the second target two-dimensional time series trajectory at the second current time node is used as the difference evaluation of water and salt lag accumulation. The absolute values ​​of the water and salt lag accumulation corresponding to the first target two-dimensional time series trajectory at the first current time node and the absolute values ​​of the water and salt lag accumulation corresponding to the second target two-dimensional time series trajectory at the second current time node are added together, and the result is added to a preset minimum positive constant. The corresponding calculation result is used as the normalization benchmark for water and salt lag accumulation. Using the difference assessment of water-salt lag accumulation as the numerator and the normalized benchmark of water-salt lag accumulation as the denominator, the resulting fraction is used as the relative difference assessment of water-salt lag accumulation. This relative difference assessment is then subjected to exponential mapping with the natural constant as the base, yielding the first optimization factor corresponding to the first target two-dimensional time-series trajectory at the first current time node and the second target two-dimensional time-series trajectory at the second current time node.

[0076] In one implementation, assume the first The two-dimensional time-series trajectory at the... The water change data for each time point corresponding to the adjacent time points are as follows: ;No. The two-dimensional time-series trajectory at the... The salinity change data for adjacent time points corresponding to each time point are: Then the first The second-order time trajectory is in the... The moment to the The expression for calculating the two-dimensional state-space path length of a time interval at time n is:

[0077]

[0078] in, Indicates the first The second-order time trajectory is in the... The moment to the The two-dimensional state-space path length of a time interval at each instant; Indicates the first The two-dimensional time-series trajectory at the... Moisture change data for each adjacent time point; Indicates the first The two-dimensional time-series trajectory at the... Salt content change data corresponding to adjacent time points at each time point.

[0079] Furthermore, the first The second-order time trajectory is in the... The formula for calculating the lag accumulation of water and salt at each time point is:

[0080]

[0081] in, Indicates the first The second-order time trajectory is in the... The lag accumulation of water and salt at each time point; Indicates the first The second-order time trajectory is in the... The moment to the The two-dimensional state-space path length of a time interval at each instant; Indicates the first The two-dimensional time-series trajectory at the... Moisture change data for each adjacent time point; Indicates the first The two-dimensional time-series trajectory at the... Salt content change data corresponding to adjacent time points at each time point.

[0082] Furthermore, the first The two-dimensional time-series trajectory at the... The time node and the first The two-dimensional time-series trajectory at the... The expression for calculating the first optimization factor corresponding to each time node is:

[0083]

[0084] in, Indicates the first The two-dimensional time-series trajectory at the... The time node and the first The two-dimensional time-series trajectory at the... The first optimization factor corresponding to each time point; Indicates the first The second-order time trajectory is in the... The lag accumulation of water and salt at each time point; Indicates the first The second-order time trajectory is in the... The lag accumulation of water and salt at each time point; This indicates absolute value calculation; This represents a preset, extremely small positive constant, which is set in the embodiments of the present invention. .

[0085] It should be noted that, firstly, considering that the hysteresis effect is essentially the influence of historical water changes on subsequent salinity changes over a long period, the formula employs a double-layer summation structure. The outer summation iterates through all water fluctuations from the initial point to the current node, while the inner summation accumulates all salinity changes since the occurrence of the current water fluctuation. The product of these two summations effectively measures the direct driving force of water supply at a specific historical moment on subsequent salinity discharge. Secondly, in real farmland soil environments, the driving effect of early water replenishment on current salinity changes continuously decays. Existing technologies that use fixed time windows or constant decay rates to constrain this effect cannot adapt to complex and variable soil environments. Therefore, this formula introduces the cumulative evolution path length in a two-dimensional state space as the denominator for natural decay. The longer the distance the water and salt states travel in phase space, the longer the physical replacement process the plot has undergone, and the weaker the driving relationship between early water fluctuations and current salinity. Finally, during the algorithm optimization process, when attempting to align and match data points from a fast-permeable plot with concurrent data points from a slow-permeable plot, the delayed accumulation of water and salt in the two plots will exhibit a significant difference. The formula utilizes an exponential structure to handle this relative difference, transforming the originally hidden physical periodic misalignment into a substantial numerical penalty coefficient. Under the mechanism of dynamic time warping to find the global minimum cumulative distance, this dramatically amplified local alignment cost will directly block the ill-conditioned warping path of the algorithm, forcing it to only align with data points that have evolved to a similar lag stage. This fundamentally solves the problem of misjudging the plot evolution mechanism caused by existing algorithms ignoring the asynchronous lag characteristics of water and salt.

[0086] Step S3: Obtain the second optimization factor by performing local water-salt trajectory fluctuation analysis on the two-dimensional time-series trajectory set.

[0087] After introducing a first optimization factor to constrain the ill-conditioned stretching alignment of the hysteresis cycle, when strong interventions like flood irrigation and weak interventions like micro-drip irrigation occur, although the absolute amplitudes of their water-salt dynamic responses differ greatly, their inherent hysteresis structure—where rising water leads to falling salt—i.e., the water-salt ratio response relationship, may be similar. In this case, the water-salt lag accumulation calculated from the trajectories of these two types of plots will show similar ratios, causing the first optimization factor to approach 1 and become ineffective in applying penalty. However, the water-salt interaction caused by flood irrigation is extremely drastic, leaving numerous high-frequency abrupt changes and sharp inflections on its time-series trajectory; while the water-salt changes caused by micro-drip irrigation exhibit extremely smooth perturbation curves. Existing dynamic time warping algorithms lack the ability to perceive differences in the kinetic energy of local evolution in sequences when accumulating local distances, easily accumulating and forcibly aligning the locally violent oscillation waveforms under strong intervention with the locally smooth perturbation waveforms under weak intervention as small distances. This smoothing misjudgment, ignoring the intensity of fluctuations, leads to the system's inability to accurately distinguish the magnitude of abrupt energy changes during water-salt transport. Therefore, the solution in this step is as follows: after confirming that the two trajectory points are in similar hysteresis stages, it is necessary to further examine whether the intensity of water and salt evolution experienced by the two is consistent. It is necessary to eliminate the interference of absolute numerical values ​​of sensor dimensions and instead use the degree of local curvature and reversal of the path swept by the trajectory in the two-dimensional state space to quantify its endogenous evolutionary kinetic energy, and apply a second layer of penalty correction to the alignment operation with significant differences in fluctuation intensity.

[0088] In summary, this invention first extracts local fluctuation features from moisture and salinity data in a two-dimensional time-series trajectory set to obtain local water-salt trajectory fluctuation feature data. Specifically, for any first target two-dimensional time-series trajectory to be clustered, the original soil moisture and soil salinity data corresponding to the first target two-dimensional time-series trajectory are extracted from the two-dimensional time-series trajectory set. For any first current time node in the first target two-dimensional time-series trajectory, when the first current time node has original soil moisture and soil salinity data corresponding to the previous and second previous time nodes, the original soil moisture data corresponding to the first current time node is combined with twice the original soil moisture data corresponding to the previous time node and the original soil moisture data corresponding to the second previous time node to obtain the second-order moisture fluctuation assessment corresponding to the first current time node. The original soil salinity data corresponding to the first current time node is combined with twice the original soil salinity data corresponding to the previous time node and the original soil salinity data corresponding to the second previous time node to obtain the second-order salinity fluctuation assessment corresponding to the first current time node. The square root of the sum of the squares of the second-order moisture fluctuation assessment and the square root of the square of the second-order salinity fluctuation assessment at the first current time node is used to obtain the local fluctuation intensity assessment at the first current time node. For any first current time node in the first target two-dimensional time-series trajectory, all acquired local fluctuation intensity assessments are accumulated to obtain the local water and salinity trajectory fluctuation characteristic data of the first target two-dimensional time-series trajectory at the first current time node. For any second target two-dimensional time-series trajectory used for comparison and matching, the same processing method as for the first target two-dimensional time-series trajectory is used to obtain the local water and salinity trajectory fluctuation characteristic data of the second target two-dimensional time-series trajectory at any second current time node.

[0089] After obtaining the local water and salt trajectory fluctuation characteristic data, the local water and salt trajectory fluctuation characteristic data is further normalized to obtain the local water and salt trajectory fluctuation degree. Specifically, for any first current time node in any first target two-dimensional time series trajectory to be aligned and any second current time node in any second target two-dimensional time series trajectory, the local water and salt trajectory fluctuation characteristic data corresponding to the first target two-dimensional time series trajectory at the first current time node and the local water and salt trajectory fluctuation characteristic data corresponding to the second target two-dimensional time series trajectory at the second current time node are obtained. For any first current time node in the first target two-dimensional time series trajectory, the original soil moisture data and original soil salinity data corresponding to the first current time node are extracted from the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time series trajectory, and the original soil moisture data and original soil salinity data corresponding to the starting time node are also extracted. The square root of the square of the sum of the square of the difference between the original soil moisture data corresponding to the first current time node and the original soil moisture data corresponding to the starting time node is taken to obtain the net displacement assessment of the first target two-dimensional time series trajectory at the first current time node. Using the local water-salt trajectory fluctuation characteristic data corresponding to the first target's two-dimensional time-series trajectory at the first current time node as the numerator, and the net displacement assessment of the first target's two-dimensional time-series trajectory at the first current time node as the denominator, the local water-salt trajectory fluctuation degree of the first target's two-dimensional time-series trajectory at the first current time node is obtained. For any second current time node in the second target's two-dimensional time-series trajectory, the same processing method as for the first target's two-dimensional time-series trajectory is adopted to obtain the net displacement assessment of the second target's two-dimensional time-series trajectory at the second current time node. Using the local water-salt trajectory fluctuation characteristic data corresponding to the second target's two-dimensional time-series trajectory at the second current time node as the numerator, and the net displacement assessment of the second target's two-dimensional time-series trajectory at the second current time node as the denominator, the local water-salt trajectory fluctuation degree of the second target's two-dimensional time-series trajectory at the second current time node is obtained.

[0090] After obtaining the local water-salt trajectory volatility, a second optimization factor is obtained by applying a difference penalty to the local water-salt trajectory volatility. Specifically, for any first current time node in any first target two-dimensional time-series trajectory to be aligned and any second current time node in any second target two-dimensional time-series trajectory, the local water-salt trajectory volatility corresponding to the first target two-dimensional time-series trajectory at the first current time node and the local water-salt trajectory volatility corresponding to the second target two-dimensional time-series trajectory at the second current time node are obtained. The absolute value of the difference between the local water-salt trajectory volatility corresponding to the first target two-dimensional time-series trajectory at the first current time node and the local water-salt trajectory volatility corresponding to the second target two-dimensional time-series trajectory at the second current time node is used as the local water-salt trajectory volatility difference assessment. The larger of the local water-salt trajectory volatility corresponding to the first target two-dimensional time-series trajectory at the first current time node and the local water-salt trajectory volatility corresponding to the second target two-dimensional time-series trajectory at the second current time node is added to a preset minimum positive constant, and the corresponding calculation result is used as the local water-salt trajectory volatility normalization benchmark. Using the local water-salt trajectory fluctuation difference assessment as the numerator and the local water-salt trajectory fluctuation normalization benchmark as the denominator, the resulting fraction is used as the local water-salt trajectory fluctuation relative difference assessment. Adding the local water-salt trajectory fluctuation relative difference assessment to a constant 1 yields the second optimization factor corresponding to the first target two-dimensional time series trajectory at the first current time node and the second target two-dimensional time series trajectory at the second current time node.

[0091] In one implementation, assume the first The two-dimensional time-series trajectory at the... The second-order fluctuation of moisture at each moment is assessed as follows: ;No. The two-dimensional time-series trajectory at the... The second-order fluctuation of salinity at each moment is assessed as follows: ;No. The water content of the two-dimensional time-series trajectories at the initial time is... ;No. The salt fraction value of each two-dimensional time-series trajectory at the initial time is Then the first The two-dimensional time-series trajectory at the... The expression for calculating the local water-salt trajectory fluctuation corresponding to each time point is:

[0092]

[0093] in, Indicates the first The two-dimensional time-series trajectory at the... Local water-salt trajectory fluctuations corresponding to each time point; Indicates the first The two-dimensional time-series trajectory at the... Assessment of second-order fluctuations in moisture content at each moment; Indicates the first The two-dimensional time-series trajectory at the... Assessment of second-order fluctuations in salinity at each moment; Indicates the first Moisture content of a two-dimensional time-series trajectory at the initial time; Indicates the first The salinity values ​​of a two-dimensional time-series trajectory at the initial time; Indicates the first The two-dimensional time-series trajectory at the... Moisture content at a given moment; Indicates the first The two-dimensional time-series trajectory at the... The salinity value at each moment.

[0094] Furthermore, the first The two-dimensional time-series trajectory at the... The time node and the first The two-dimensional time-series trajectory at the... The expression for calculating the second optimization factor corresponding to each time node is:

[0095]

[0096] in, Indicates the first The two-dimensional time-series trajectory at the... Local water-salt trajectory fluctuations corresponding to each time point; Indicates the first The two-dimensional time-series trajectory at the... Local water-salt trajectory fluctuations corresponding to each time point; This represents a preset, extremely small positive constant, which is set in the embodiments of the present invention. .

[0097] It should be noted that, firstly, to accurately measure the evolution intensity of water-salt interaction without relying on external parameters, in farmland monitoring, the first-order difference only represents the rate of change of state, while the second-order difference represents the acceleration of state reversal and reversal. Under heavy irrigation, water-salt interaction is intense, and the trajectory curve has many sharp inflections. At this time, the second-order difference value is extremely large, causing the numerator to accumulate rapidly and form a high fluctuation energy pool. In contrast, for curves with gentle and natural evolution, the rate of change is weak, the second-order difference is minimal, and the accumulated value of the numerator tends to be flat. Secondly, to eliminate errors caused by differences in initial water and salinity or differences in the absolute dimensions of sensors between different plots, the denominator of the formula introduces the overall net displacement in two-dimensional phase space from the starting point of the sequence to the current point as a normalization term. The ratio of cumulative acceleration to net displacement makes the value of local water-salt trajectory fluctuation a dimensionless, purely morphological feature, which can adaptively reflect how much redundant tortuous oscillation a trajectory undergoes in order to reach the final state. Finally, when aligning point pairs in the algorithm, if attempting to forcibly align high-fluctuation segments (extremely large local water-salt trajectory fluctuations) caused by flooding with low-fluctuation segments (extremely small local water-salt trajectory fluctuations) caused by minute seepage, the formula constructs an adaptive relative difference range ratio by calculating the absolute difference in fluctuation between the two and dividing by the larger of the two. This structure enables... A penalty coefficient greater than 1 is generated. When the abrupt change in the two trajectories is significantly different, this multiplicative penalty factor will linearly amplify the local distance cost of a single step, thereby effectively plugging the alignment loophole left by the first optimization factor, which has the same proportional lag but different intervention intensities. This allows the system to ultimately accurately isolate and cluster water-salt evolution blocks driven by different intervention intensities in the real physical world.

[0098] Thus, the second optimization factor was obtained by performing local water-salt trajectory fluctuation analysis on a two-dimensional time-series trajectory set.

[0099] Step S4: Obtain the final temporal similarity distance by jointly correcting the original single-step distance based on the first optimization factor and the second optimization factor.

[0100] The core idea of ​​existing Dynamic Time Warping (DTW) algorithms is to find the minimum cumulative distance path between two time series using dynamic programming. Its foundation is the computation of trajectories. Upper Points and Trajectories Upper The single-step Euclidean distance of each point is used, and then a globally optimal alignment path from the starting point to the ending point is found based on the state transition equation. Since the existing single-step Euclidean distance only represents the instantaneous numerical error between points, the global optimization process is prone to ill-conditioned stretching that deviates from actual farmland phenomena. This step uses the first optimization factor extracted in step S2. The second factor extracted in step S3 This applies to the single-step measurement stage of the original algorithm. Through the multiplicative intervention of these two factors, when the algorithm attempts to align data points that do not conform to the water-salt lag pattern of farmland or have significantly different evolutionary kinetic energies, its single-step computational cost is amplified exponentially. This reshaping of the underlying distance cost forces the dynamic programming state transition process to autonomously avoid paths that are merely numerically similar but physically misaligned when constructing the cumulative distance matrix. Ultimately, the globally cumulative distance calculated through backtracking accurately reflects the overall difference in water-salt evolution mechanisms between the two farmlands.

[0101] Specifically, for any first current time node in any first target two-dimensional time-series trajectory to be aligned and any second current time node in any second target two-dimensional time-series trajectory, the original soil moisture and soil salinity data corresponding to the first current time node are extracted from the original soil moisture and soil salinity data corresponding to the first target two-dimensional time-series trajectory, and the original soil moisture and soil salinity data corresponding to the second current time node are extracted from the original soil moisture and soil salinity data corresponding to the second target two-dimensional time-series trajectory. The square root of the square of the difference between the original soil moisture data corresponding to the first current time node and the original soil moisture data corresponding to the second current time node is obtained by adding the square of the difference between the original soil salinity data corresponding to the first current time node and the original single-step distance between the first target two-dimensional time-series trajectory at the first current time node and the original single-step distance between the second target two-dimensional time-series trajectory at the second current time node. The product of the original single-step distance and the first optimization factor and the second optimization factor is used as the corrected single-step distance between the first target two-dimensional time-series trajectory at the first current time node and the original single-step distance between the second target two-dimensional time-series trajectory at the second current time node. For any first current time node in the first target's two-dimensional temporal trajectory and any second current time node in the second target's two-dimensional temporal trajectory, the smaller of the corrected single-step distance, the minimum cumulative distance corresponding to the previous time node of the first target's two-dimensional temporal trajectory and the second current time node, the minimum cumulative distance corresponding to the previous time node of the first target's two-dimensional temporal trajectory and the previous time node of the second target's two-dimensional temporal trajectory, is added to obtain the minimum cumulative distance between the first target's two-dimensional temporal trajectory at the first current time node and the second target's two-dimensional temporal trajectory at the second current time node. The minimum cumulative distance between the end time node of the first target's two-dimensional temporal trajectory and the end time node of the second target's two-dimensional temporal trajectory is taken as the final temporal similarity distance between the two target's two-dimensional temporal trajectories.

[0102] Thus, the process of obtaining the final temporal similarity distance by jointly correcting the original single-step distance based on the first and second optimization factors is complete.

[0103] Step S5 involves clustering and extracting patterns from the final temporal similarity distance to obtain the evolution patterns of farmland water and salt.

[0104] After obtaining the final temporal similarity distance between each pair of monitored plots, a temporal distance matrix corresponding to all monitored plots within the target farmland area is constructed. This temporal distance matrix is ​​then input into a clustering model to perform clustering processing, thereby obtaining the cluster category label corresponding to each monitored plot. In specific implementation, the k-medoids clustering algorithm can be used, with the final temporal similarity distance as the basis for measuring the similarity between plots. By iteratively updating the cluster center plots and continuously adjusting the category of each monitored plot, the cluster center no longer changes or a preset iteration termination condition is reached, thus completing the clustering of all monitored plots.

[0105] After completing the clustering, two-dimensional time-series trajectories of the corresponding cluster center plots are extracted for each cluster category, or statistical analysis is performed on the two-dimensional time-series trajectories of all monitored plots within the same cluster category to obtain typical two-dimensional time-series trajectories characterizing the water-salt evolution features of that type of plot. Furthermore, by combining the amplitude of moisture fluctuations, the amplitude of salinity response, and the lag relationship between moisture and salinity changes in the typical two-dimensional time-series trajectories, the water-salt evolution characteristics of farmland in the corresponding cluster categories are summarized to obtain the patterns of farmland water-salt evolution. Based on practical application needs, the patterns of farmland water-salt evolution can be categorized as strong intervention and intense leaching, weak intervention and slow seepage, or natural evaporation and salinization, etc., and the corresponding plot classification results are output, providing a basis for subsequent farmland irrigation regulation, salt drainage management, and zoning management.

[0106] Thus, the study completed the acquisition of farmland water and salt evolution patterns by clustering and extracting patterns from the final temporal similarity distance.

[0107] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for mining farmland water-salinity evolution patterns based on temporal trajectory clustering, characterized in that, The system includes: Step S1: Obtain a two-dimensional time-series trajectory set by collecting and preprocessing raw data on farmland moisture and salinity; Step S2: Obtain the first optimization factor by performing water-salt hysteresis cumulative effect analysis on the two-dimensional time-series trajectory set; Step S3: Obtain the second optimization factor by performing local water-salt trajectory fluctuation analysis on the two-dimensional time-series trajectory set; Step S4: Obtain the final temporal similarity distance by jointly correcting the original single-step distance based on the first optimization factor and the second optimization factor; Step S5: Obtain the evolution pattern of farmland water and salt by clustering and extracting patterns from the final temporal similarity distance; The step of obtaining the first optimization factor by performing water and salt lag accumulation effect analysis on the two-dimensional time-series trajectory set includes: obtaining the water and salt lag accumulation amount by performing correlation accumulation processing on the moisture change data and salt change data in the two-dimensional time-series trajectory set; and obtaining the first optimization factor by performing difference penalty processing on the water and salt lag accumulation amount. The step of obtaining the second optimization factor by performing local water and salt trajectory fluctuation analysis on a two-dimensional time-series trajectory set includes: extracting local fluctuation features from the moisture and salinity data in the two-dimensional time-series trajectory set to obtain local water and salt trajectory fluctuation feature data; performing normalization calculation on the local water and salt trajectory fluctuation feature data to obtain local water and salt trajectory fluctuation degree; and performing difference penalty processing on the local water and salt trajectory fluctuation degree to obtain the second optimization factor. The method of obtaining the final temporal similarity distance by jointly correcting the original single-step distance based on the first optimization factor and the second optimization factor includes: extracting the original soil moisture data and original soil salinity data of the two-dimensional temporal trajectory to be aligned at each time node; using the product of the original single-step distance and the first optimization factor and the second optimization factor as the corrected single-step distance between the first target two-dimensional temporal trajectory at the first current time node and the second target two-dimensional temporal trajectory at the second current time node; based on this, constructing the minimum cumulative distance of dynamic time warping based on the corrected single-step distance between each time node, and using the minimum cumulative distance corresponding to the last time node of the two-dimensional temporal trajectory to be aligned as the final temporal similarity distance.

2. The system for mining farmland water-salt evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The process of acquiring a two-dimensional time-series trajectory set by collecting and preprocessing raw data on farmland moisture and salinity includes: A soil sensing sensor network is deployed in the target farmland area according to a preset spatial grid. Soil moisture sensor nodes for collecting raw soil moisture data and soil conductivity sensor nodes for collecting raw soil salinity data are set at the monitoring plots corresponding to each spatial grid. A fixed sampling frequency is set, and each monitoring plot is continuously monitored and collected according to the fixed sampling frequency. The collection time covers a complete irrigation and evaporation cycle to obtain the raw soil moisture data and raw soil conductivity data corresponding to each monitoring plot. Based on the preset empirical conversion relationship between soil electrical conductivity and salinity, the raw soil electrical conductivity data of each monitoring plot are processed by salinity conversion to obtain the raw soil salinity data of each monitoring plot; wherein, the raw soil moisture data is the soil volumetric water content time series data, and the raw soil salinity data is the soil total salinity time series data. Outlier removal, missing value completion, and time axis alignment were performed on the raw soil moisture and soil salinity data of each monitoring plot. The processed raw soil moisture and soil salinity data were then fused to obtain a set of two-dimensional time-series trajectories for each monitoring plot.

3. The system for mining farmland water-salt evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The step of obtaining the water-salinity lag accumulation by correlating and accumulating the moisture change data and salinity change data in the two-dimensional time-series trajectory set includes: For any first target two-dimensional time-series trajectory to be clustered, the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory are extracted from the set of two-dimensional time-series trajectories. Then, the first-order difference calculation of adjacent time points is performed on the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory to obtain the moisture change data and salinity change data corresponding to the first target two-dimensional time-series trajectory. For any first historical time node and any first current time node in the two-dimensional time-series trajectory of the first target, when the first current time node is not earlier than the first historical time node, the moisture change data corresponding to the first historical time node is used as the first moisture-driven assessment; for each time node between the first historical time node and the first current time node, the salinity change data corresponding to each time node is used as the corresponding salinity response assessment. For any target time interval between the first historical time node and the first current time node, the square root of the sum of the squares of the moisture change data and the squares of the salinity change data at adjacent moments within the target time interval is taken, and the square root results corresponding to all adjacent moments are accumulated. The accumulated result is taken as the two-dimensional state space path length between the first historical time node and the first current time node. The salinity response assessment corresponding to each time node is divided by the sum of a constant and the two-dimensional state space path length, and the calculated results are accumulated to obtain the cumulative salinity decay response assessment. The result of multiplying the first moisture driving assessment and the cumulative salinity decay response assessment is taken as the hysteresis driving contribution corresponding to the first historical time node. For the first current time node, the lag driving contribution of all the first historical time nodes before the first current time node is accumulated to obtain the water and salt lag accumulation of the first target two-dimensional time series trajectory at the first current time node. For any second target two-dimensional time-series trajectory used for comparison and matching, the same processing method as the first target two-dimensional time-series trajectory is adopted to obtain the water and salt lag accumulation corresponding to any second current time node of the second target two-dimensional time-series trajectory.

4. The system for mining farmland water-salt evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The step of obtaining the first optimization factor by applying a difference penalty to the accumulated water and salt lag includes: For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, obtain the water and salt lag accumulation corresponding to the first target two-dimensional time trajectory at the first current time node and the water and salt lag accumulation corresponding to the second target two-dimensional time trajectory at the second current time node. The absolute value of the difference between the water and salt lag accumulation of the first target two-dimensional time series trajectory at the first current time node and the water and salt lag accumulation of the second target two-dimensional time series trajectory at the second current time node is used as the water and salt lag accumulation difference assessment. The absolute value of the water and salt lag accumulation of the first target two-dimensional time trajectory at the first current time node is added to the absolute value of the water and salt lag accumulation of the second target two-dimensional time trajectory at the second current time node, and the addition result is added to a preset minimum positive constant. The corresponding calculation result is used as the normalization benchmark for the water and salt lag accumulation. The difference assessment of water and salt lag accumulation is used as the numerator, the normalized benchmark of water and salt lag accumulation is used as the denominator, and the corresponding fraction is used as the relative difference assessment of water and salt lag accumulation. The relative difference assessment of water and salt lag accumulation is processed by exponential mapping with the natural constant as the base, and the first optimization factor corresponding to the first target two-dimensional time series trajectory at the first current time node and the second target two-dimensional time series trajectory at the second current time node is obtained.

5. The system for mining farmland water-salt evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The process of extracting local fluctuation features from moisture and salinity data in a two-dimensional time-series trajectory set to obtain local water-salt trajectory fluctuation feature data includes: For any first target two-dimensional time-series trajectory to be clustered, extract the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory from the set of two-dimensional time-series trajectories; For any first current time node in the two-dimensional time-series trajectory of the first target, when the first current time node has the original soil moisture data and original soil salinity data corresponding to the previous time node and the two previous time nodes, the original soil moisture data corresponding to the first current time node is combined with twice the original soil moisture data corresponding to the previous time node and the original soil moisture data corresponding to the two previous time nodes to obtain the second-order moisture fluctuation assessment corresponding to the first current time node; the original soil salinity data corresponding to the first current time node is combined with twice the original soil salinity data corresponding to the previous time node and the original soil salinity data corresponding to the two previous time nodes to obtain the second-order salinity fluctuation assessment corresponding to the first current time node. The square root of the sum of the square of the second-order fluctuation assessment of moisture and the square of the second-order fluctuation assessment of salinity at the first current time node is used to obtain the local fluctuation intensity assessment at the first current time node. For any first current time node in the two-dimensional time-series trajectory of the first target, all the acquired local fluctuation intensity assessments are accumulated to obtain the local water and salt trajectory fluctuation characteristic data corresponding to the first current time node of the two-dimensional time-series trajectory of the first target. For any second target two-dimensional time-series trajectory used for comparison and matching, the same processing method as the first target two-dimensional time-series trajectory is adopted to obtain the local water and salt trajectory fluctuation feature data corresponding to any second current time node of the second target two-dimensional time-series trajectory.

6. The system for mining farmland water-salt evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The process of obtaining the local water-salt trajectory fluctuation degree by normalizing the local water-salt trajectory fluctuation characteristic data includes: For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, obtain the local water and salt trajectory fluctuation feature data corresponding to the first target two-dimensional time trajectory at the first current time node and the local water and salt trajectory fluctuation feature data corresponding to the second target two-dimensional time trajectory at the second current time node; For any first current time node in the first target two-dimensional time-series trajectory, extract the original soil moisture data and original soil salinity data corresponding to the first current time node from the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time-series trajectory, and extract the original soil moisture data and original soil salinity data corresponding to the starting time node; add the square of the difference between the original soil moisture data corresponding to the first current time node and the original soil moisture data corresponding to the starting time node, and take the square root of the sum of the square of the difference between the original soil salinity data corresponding to the first current time node and the original soil salinity data corresponding to the starting time node to obtain the net displacement assessment of the first target two-dimensional time-series trajectory corresponding to the first current time node; Using the local water and salt trajectory fluctuation characteristic data of the first target two-dimensional time-series trajectory at the first current time node as the numerator and the net displacement assessment of the first target two-dimensional time-series trajectory at the first current time node as the denominator, the local water and salt trajectory fluctuation degree of the first target two-dimensional time-series trajectory at the first current time node is obtained. For any second current time node in the second target two-dimensional time-series trajectory, the same processing method as for the first target two-dimensional time-series trajectory is adopted to obtain the net displacement assessment of the second target two-dimensional time-series trajectory at the second current time node. The local water and salt trajectory fluctuation characteristic data of the second target two-dimensional time-series trajectory at the second current time node is used as the numerator, and the net displacement assessment of the second target two-dimensional time-series trajectory at the second current time node is used as the denominator to obtain the local water and salt trajectory fluctuation degree of the second target two-dimensional time-series trajectory at the second current time node.

7. The system for mining farmland water-salt evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The process of obtaining a second optimization factor by applying a difference penalty to the local water-salt trajectory fluctuation includes: For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, obtain the local water and salt trajectory fluctuation of the first target two-dimensional time trajectory at the first current time node and the local water and salt trajectory fluctuation of the second target two-dimensional time trajectory at the second current time node. The absolute value of the difference between the local water and salt trajectory fluctuation of the first target two-dimensional time-series trajectory at the first current time node and the local water and salt trajectory fluctuation of the second target two-dimensional time-series trajectory at the second current time node is used as the local water and salt trajectory fluctuation difference assessment. The larger of the local water and salt trajectory fluctuation of the first target two-dimensional time-series trajectory at the first current time node and the local water and salt trajectory fluctuation of the second target two-dimensional time-series trajectory at the second current time node is added to a preset minimum positive constant, and the corresponding calculation result is used as the normalization benchmark for local water and salt trajectory fluctuation. The local water-salt trajectory fluctuation difference assessment is used as the numerator, the local water-salt trajectory fluctuation normalization benchmark is used as the denominator, and the corresponding fraction is used as the local water-salt trajectory fluctuation relative difference assessment. The relative difference assessment of local water-salt trajectory fluctuation is added to a constant 1 to obtain the second optimization factor corresponding to the first target two-dimensional time-series trajectory at the first current time node and the second target two-dimensional time-series trajectory at the second current time node.

8. The system for mining farmland water-salinity evolution patterns based on time-series trajectory clustering according to claim 1, characterized in that, The process of obtaining the final temporal similarity distance by jointly correcting the original single-step distance based on a first optimization factor and a second optimization factor includes: For any first current time node in any first target two-dimensional time trajectory to be aligned and any second current time node in any second target two-dimensional time trajectory, extract the original soil moisture data and original soil salinity data corresponding to the first current time node from the original soil moisture data and original soil salinity data corresponding to the first target two-dimensional time trajectory, and extract the original soil moisture data and original soil salinity data corresponding to the second current time node from the original soil moisture data and original soil salinity data corresponding to the second target two-dimensional time trajectory. The square root of the square of the difference between the original soil moisture data corresponding to the first current time node and the original soil moisture data corresponding to the second current time node is obtained by adding the square of the difference between the original soil salinity data corresponding to the first current time node and the original soil salinity data corresponding to the second current time node. The product of the original single-step distance and the first optimization factor and the second optimization factor is used as the corrected single-step distance between the first target two-dimensional time-series trajectory at the first current time node and the second target two-dimensional time-series trajectory at the second current time node. For any first current time node in the two-dimensional time-series trajectory of the first target and any second current time node in the two-dimensional time-series trajectory of the second target, the smaller of the following values ​​is added: the correction step distance, the minimum cumulative distance corresponding to the previous time node of the first current time node and the second current time node, the minimum cumulative distance corresponding to the previous time node of the first current time node and the second current time node, and the minimum cumulative distance corresponding to the previous time node of the first current time node and the previous time node of the second current time node. This yields the minimum cumulative distance between the two-dimensional time-series trajectory of the first target at the first current time node and the two-dimensional time-series trajectory of the second target at the second current time node. The minimum cumulative distance between the end time node of the first target's two-dimensional time-series trajectory and the end time node of the second target's two-dimensional time-series trajectory is taken as the final temporal similarity distance between the two-dimensional time-series trajectories of the first and second targets.

Images