A multifunctional soil detection system

Abstract

The present invention provides a multi-functional soil detection system, which relates to the technical field of data processing. The system includes: an association verification module, configured to construct a temperature-humidity-conductivity association matrix according to standard soil data, and obtain a matching coefficient by checking the difference distribution between the first temperature value, the first humidity value, and the first conductivity value in the verification matrix; an abnormality judgment module, configured to judge whether there is an abnormal data reading at a preset detection point according to the matching coefficient. If the matching coefficient is lower than a preset stability threshold, an abnormality prompt signal is generated; a reconstruction module, configured to perform substitution calculation on abnormal sub-items in the standard soil data according to the abnormality prompt signal, and generate reconstructed soil data by using the method of curve fitting based on historical isotopic point data; an integration and output module, configured to merge the reconstructed soil data with the standard soil data without abnormalities to obtain the final soil detection data. The present invention improves the autonomy and accuracy of the soil detection system.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a multifunctional soil detection system. Background Art

[0002] In the existing technology, soil detection systems mainly collect soil data of different dimensions, such as moisture, pH value, conductivity and temperature, by deploying a variety of sensors, and then read and analyze the data through a single-chip microcomputer or embedded processor. Some systems also have wireless data transmission functions to achieve real-time monitoring of field soil conditions. Such systems are often used in agricultural production management, environmental monitoring and other scenarios to assist users in making scientific fertilization, irrigation and other decisions. During the implementation process, some devices are also equipped with GPS modules to achieve geographic positioning of data and improve the spatial accuracy of soil information.

[0003] However, in the horticultural greenhouse management scenario, the existing soil detection system often needs to deploy different sensor modules separately due to its low functional integration, and access the collection equipment through multiple interfaces, resulting in a complex system structure. This not only increases the difficulty of operation during the deployment process, but also easily causes data reading anomalies due to poor interface contact. For example, in a high humidity environment, the interface part of the soil temperature sensor is prone to oxidation, resulting in intermittent loss of temperature data, which in turn affects the accuracy of the automatic temperature control strategy based on temperature changes. This reliability problem is more obvious in a closed environment, limiting the system's ability to operate stably in the long term. Summary of the invention

[0004] The object of the present invention is to provide a multifunctional soil detection system, aiming to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] A multifunctional soil detection system, the system comprising:

[0007] The correlation verification module is used to construct a temperature-humidity-conductivity correlation matrix according to the standard soil data, and obtain a set of matching coefficients by verifying the difference distribution between the first temperature value and the first humidity value and the first conductivity value in the matrix;

[0008] An abnormality judgment module is used to judge whether there is a data reading abnormality at a preset detection point according to the matching coefficient, and generate an abnormality prompt signal if the matching coefficient is lower than a preset stability threshold;

[0009] The reconstruction module is used to perform alternative calculations on the abnormal sub-items in the standard soil data according to the abnormal prompt signal, and generate reconstructed soil data by curve fitting based on historical co-site data;

[0010] An integration output module is used to merge the reconstructed soil data with the standard soil data without anomalies to obtain the final soil test data, and package the final soil test data correspondingly and output it to the backend processing module.

[0011] Preferably, the association verification module includes:

[0012] A first modeling sub-module is used to construct a corresponding three-dimensional space point set according to the first temperature value, the first humidity value, and the first conductivity value in the standard soil data, and respectively fit and generate three groups of association plane equations for the temperature-humidity plane, the electric-temperature plane, and the electric-humidity plane on the three-dimensional space point set;

[0013] A similarity calculation sub-module is used to map the standard soil data onto the association plane equations, calculate the normal distance between the mapped points and the association planes, and obtain three groups of spatial offset values;

[0014] A weight fusion sub-module is used to construct a temperature-humidity-conductivity association matrix according to the three groups of spatial offset values, perform normalization processing on it, extract the difference distribution characteristics between the parameters, and calculate the target matching coefficient.

[0015] Preferably, the reconstruction module includes:

[0016] A time reversal sub-module is used to, after the anomaly prompt signal is activated, call the standard soil data within the nearest preset number of anomaly-free detection cycles of the preset detection point in the historical database, and arrange the data in reverse order to generate a time-series reversed data set;

[0017] A trend prediction sub-module is used to construct a first-order state transition equation according to the historical values corresponding to the anomaly sub-items in the time-series reversed data set, combine the known non-anomaly sub-items, and use cross-regression to perform trend fitting on the missing values to generate predicted interpolations;

[0018] A structure constraint sub-module is used to introduce an upper tolerance constraint of the matching coefficient, perform error feedback correction on the predicted interpolations, eliminate the interpolation points that exceed the preset dynamic tolerance threshold range during the fitting process, and generate the reconstructed soil data.

[0019] Preferably, the first modeling sub-module includes:

[0020] A window screening unit is used to extract multiple data nodes within a preset time window according to the standard soil data to form a three-dimensional space point set for modeling;

[0021] A plane fitting unit is used to perform fitting processing on the three-dimensional space point set to generate three groups of association plane equations for the temperature-humidity plane, the electric-temperature plane, and the electric-humidity plane;

[0022] A reference plane calibration unit for dynamically determining the initial normal vector direction of each group of planes according to the statistical results of the historical point set distributions of different detection points.

[0023] Preferably, the similarity calculation sub-module includes:

[0024] A projection transformation unit for mapping the standard soil data in the current detection period to spatial detection points, projecting them onto the corresponding planes in the associated plane equation, and obtaining projection point data;

[0025] A distance calculation unit for respectively calculating the temperature-humidity offset value, the temperature-electricity offset value, and the electricity-humidity offset value according to the normal perpendicular distance between the projection points and the original spatial detection points, and forming a set of spatial offset values;

[0026] An offset adjustment unit for abnormally marking the offset values in the set of spatial offset values that fluctuate beyond a preset stability threshold.

[0027] Preferably, the weight fusion sub-module includes:

[0028] A weight assignment unit for assigning weighting coefficients to the temperature-humidity offset value, the temperature-electricity offset value, and the electricity-humidity offset value according to the historical stability scores of the sensors corresponding to the preset detection points, generating an initial offset matrix, and each column of the initial offset matrix corresponds to three groups of spatial offset values;

[0029] A difference extraction unit for performing normalization processing on the initial offset matrix, constructing a temperature-humidity correlation matrix, and extracting the difference distribution characteristics between different parameter combinations in the matrix, and the difference distribution characteristics are calculated based on the covariance distance between the parameters;

[0030] A matching calculation unit for inputting the difference distribution characteristics into a preset weighted scoring model and calculating the target matching coefficient in the current detection period.

[0031] Preferably, the time reversal sub-module includes:

[0032] A historical screening unit for screening data segments in the historical database with a similarity higher than a similarity threshold to the environmental parameters in the current detection period within the last three anomaly-free detection periods, and generating an inversion reference data source;

[0033] A time sequence sorting unit for sorting the standard soil data in the inversion reference data source in reverse order according to the acquisition timestamps, and maintaining the corresponding relationships of the first temperature value, the first humidity value, and the first conductivity value, and constructing a preliminary time sequence inversion data set;

[0034] A boundary cleaning unit for performing extreme value identification and elimination operations on the boundary points in the preliminary time sequence inversion data set, and generating a time sequence inversion data set.

[0035] Preferably, the trend prediction sub-module includes:

[0036] A state modeling unit, configured to construct a first-order linear state transition equation according to the historical values corresponding to the abnormal sub-items in the time series inversion dataset and introducing a time-decreasing weight to apply an attenuation factor to the long-distance historical data;

[0037] A cross-fitting unit, configured to introduce the real-time data of the non-abnormal sub-items as joint input variables into the first-order linear state transition equation, perform cross-regression modeling, and generate a preliminary prediction interpolation on the premise of maintaining the correlation of the existing variables;

[0038] A confidence evaluation unit, configured to calculate the sum of squared fitting residuals of the preliminary prediction interpolation and compare it with a preset confidence threshold. If the sum of squared residuals is lower than the confidence threshold, mark the corresponding preliminary prediction interpolation as a valid prediction value and generate a prediction interpolation.

[0039] Preferably, the structure constraint sub-module includes:

[0040] An error evaluation unit, configured to calculate a matching error score according to the difference between the prediction interpolation and the standard soil data, in combination with the sum of squared fitting residuals of the prediction interpolation;

[0041] A tolerance limit unit, configured to set a dynamic tolerance threshold based on a matching coefficient and compare the matching error score with the dynamic tolerance threshold to screen out valid interpolation points that meet the dynamic tolerance threshold range;

[0042] A feedback correction unit, configured to perform a quadratic fitting process based on the principle of residual minimization on the interpolation points that do not meet the dynamic tolerance threshold range, and merge the corrected interpolation points with the valid interpolation points to generate reconstructed soil data.

[0043] Preferably, the cross-fitting unit includes:

[0044] A variable evaluation sub-unit, configured to extract the historical values of the abnormal sub-items from the time series inversion dataset based on the first-order linear state transition equation, and use the real-time data of the non-abnormal sub-items as joint input variables to calculate the correlation coefficient and its prediction residual variance between each non-abnormal sub-item and the abnormal sub-item, and generate an influence value;

[0045] A model selection sub-unit, configured to judge the prediction effectiveness of the non-abnormal sub-items according to the influence value. If the influence value is higher than a preset influence threshold, use the non-abnormal sub-item as an input variable to construct a multi-linear state transition path; if the influence value is lower than the preset influence threshold, input it into a regression path constructed based on a support vector structure to generate two groups of candidate prediction interpolations respectively;

[0046] A prediction combinatorial subunit is used to perform weighted fusion on two sets of prediction interpolations to generate a preliminary prediction interpolation.

[0047] The above solution of the present invention has at least the following beneficial effects:

[0048] In a high-humidity and enclosed environment with high requirements for detection accuracy, such as horticultural greenhouse management, this system can effectively overcome the stability bottleneck brought by problems such as redundant sensor interfaces, multi-module deployment, and frequent data reading anomalies in existing soil detection systems. On the one hand, the acquisition module synchronously collects key parameters in the original soil data, such as the first temperature value, the first humidity value, and the first conductivity value, avoiding the cumbersome operations of separately deploying multiple sensor modules in the past, reducing the overall system layout complexity, reducing the possibility of failures caused by redundant interfaces, and enhancing the physical anti-interference ability of the detection system in a high-humidity environment.

[0049] On the other hand, by setting a preprocessing module to perform signal correction and format standardization processing on the original soil data, the output results of various sensors are fused under a unified data structure, effectively solving the problem of difficult data alignment caused by differences in sensor output standards, and enhancing the system's ability to identify and correct mutation data. Especially during the data reading process, once an abnormal jump occurs in a certain dimension parameter, the system can construct a temperature-humidity-conductivity correlation matrix through the subsequent correlation verification module, and dynamically calculate the matching coefficient using the difference distribution between parameters, so as to achieve logical identification of abnormal signals without relying on hardware redundancy.

[0050] In addition, the system introduces a stability threshold judgment mechanism by setting an abnormal judgment module, avoiding the inflexibility problem caused by judging the absolute value threshold of the original data, but judging the data stability from the multi-dimensional collaborative trend, thus being more applicable to the scenario where the soil state in the greenhouse changes slowly but has strong correlations. If an abnormality is identified, the reconstruction module will perform curve fitting on the historical data at the same position to reconstruct the soil parameters, effectively making up for the data loss caused by abnormal acquisition, and avoiding the chain reaction of affecting the overall control strategy due to local failures.

[0051] Finally, through the integration output module, the reconstructed soil data and the original standard data are fused, and the packaged detection results are output to the backend processing module, providing stable and complete data support for the greenhouse temperature control and irrigation control systems. The overall solution not only improves the system integration and data stability, but also enhances the operation robustness in a specific enclosed environment, having good engineering practical value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is an architecture diagram of a multi-functional soil detection system provided by an embodiment of the present invention. Detailed implementation manners

[0053] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0054] As Figure 1 shown, an embodiment of the present invention provides a multi-functional soil detection system, and the system includes:

[0055] An acquisition module, configured to acquire original soil data at a preset detection point, where the original soil data includes a first temperature value, a first humidity value, and a first conductivity value, and is used to reflect the soil basic state information under the current environment;

[0056] A preprocessing module, configured to perform signal correction and format standardization processing on the original soil data to obtain standard soil data, and the standard soil data is used to eliminate the discrete anomalies caused by uneven sensor layout or electrical interference;

[0057] An association verification module, configured to construct a temperature-humidity-conductivity association matrix according to the standard soil data, and obtain a set of matching coefficients by verifying the difference distribution between the first temperature value and the first humidity value and the first conductivity value in the matrix, and is used to represent the internal stability of the soil information at the preset detection point;

[0058] An anomaly judgment module, configured to judge whether there is data reading anomaly at the preset detection point according to the matching coefficient. If the matching coefficient is lower than the preset stability threshold, an anomaly prompt signal is generated, otherwise the monitoring process continues to execute;

[0059] A reconstruction module, configured to perform substitution calculation on the abnormal sub-items in the standard soil data on the basis of generating the anomaly prompt signal, and generate reconstructed soil data by using the curve fitting method based on the historical isotopic point data, and is used to restore the complete data set of the current detection point;

[0060] An integration and output module, configured to merge the reconstructed soil data with the standard soil data that has not undergone anomalies to obtain the final soil detection data, and pack the final soil detection data correspondingly and output it to the backend processing module for subsequent execution of temperature control or irrigation strategy decision-making.

[0061] In the embodiments of the present invention, in practical applications, for the multi-source complex environment at the soil detection site, single-dimensional collection and analysis often easily lead to problems such as incomplete measurement, large signal interference, and fuzzy defect positioning, making it difficult to support the continuous perception requirements of the dynamic state of the soil in scenarios such as high-precision agriculture, intelligent irrigation, and ecological restoration. In order to improve the integrity, stability, and intelligent compensation ability of the detection process, this system introduces a modular design and constructs a multi-functional soil detection system with a closed-loop functional link around the whole process of original data acquisition, signal correction, parameter verification, anomaly recognition, defect reconstruction, and data integration.

[0062] First, the acquisition module directly faces the preset detection points and collects original soil data including the first temperature value, the first humidity value, and the first conductivity value through the soil embedded sensing unit, achieving multi-parameter synchronous acquisition from the source. This multi-dimensional acquisition method lays a data foundation for constructing the parameter correlation relationship subsequently.

[0063] The preprocessing module can perform noise elimination, format standardization, signal offset correction, etc. on the collected original soil data to generate standard soil data with structured attributes, ensuring the comparability and usability of data between different dimensions.

[0064] Based on the standard soil data, the correlation verification module constructs a temperature-humidity-conductivity correlation matrix and extracts the matching coefficient using the difference distribution characteristics among the three types of parameters. The system uses this to judge whether there are problems such as abnormal outliers and trend breaks between different parameters. This link avoids directly judging whether the absolute value of each parameter is abnormal, but makes a system judgment through the multi-dimensional collaborative change trend, thereby enhancing the overall perception ability of abnormal behaviors.

[0065] The anomaly judgment module relies on the stability judgment mechanism of the matching coefficient and uses the stability threshold as the benchmark for defining anomalies. If the matching coefficient is less than this threshold, it is considered that there is data anomaly at the current detection point, and an anomaly prompt signal is triggered to achieve real-time identification of local problems.

[0066] When the anomaly prompt signal is activated, the reconstruction module can call historical data and perform alternative calculations on the abnormal sub-items in the standard soil data, generating reconstructed soil data with time consistency and environmental structure constraints in the historical data structure in a curve fitting manner to achieve the purpose of restoring missing or incorrect information.

[0067] Finally, the integration output module performs fusion processing on the reconstructed data and the original data without anomalies, outputs the final soil detection data, and pushes it to the backend platform in a packaged format to achieve the transmission of soil detection information with high robustness and high availability.

[0068] Through the linkage design of the above functional modules, on the basis of maintaining a relatively simple hardware structure, the system significantly improves the sensitivity of abnormal soil data recognition, the intelligence of defect repair, and the robustness of output data, and is particularly suitable for popularization and use in soil monitoring tasks under complex climate conditions or multi-point distributed layouts.

[0069] Among them, the main function of the acquisition module is to collect the core parameters in the soil, specifically including the first temperature value, the first humidity value, and the first conductivity value. To ensure the synchronization and stability of data acquisition, the acquisition module adopts an integrated sensing probe design, integrating the three types of sensors into one structure, avoiding problems such as complex interfaces and position offsets caused by the layout of multiple devices. The overall structure of the probe is encapsulated with waterproof and anti-corrosion packaging materials, suitable for environments such as high humidity and high salt. The lower end of the probe is in direct contact with the soil through a stainless steel or resin-wrapped pin of a certain length, and stable data is collected by inserting the same depth into the soil layer.

[0070] The acquisition circuit is connected to a multi-channel analog-to-digital converter (ADC) through a unified interface board, and the data of each sensor is synchronously collected after being conditioned by the analog front-end circuit. In terms of acquisition cycle control, the acquisition module performs data reading at fixed intervals and simultaneously supports the environmental trigger mechanism (such as temperature mutation) to start temporary acquisition tasks to ensure the timeliness of the detected data. In terms of data structure, the acquisition result is formatted into a triple with a timestamp, corresponding to the first temperature value, the first humidity value, and the first conductivity value respectively, and is transmitted to the system processing unit through the bus protocol for subsequent module call and analysis.

[0071] This structure not only ensures the consistency of data sources and anti-interference ability, but also improves the layout efficiency of the module, and is suitable for application scenarios with limited space and drastic environmental changes such as horticultural greenhouses.

[0072] In a preferred embodiment of the present invention, the correlation verification module includes:

[0073] The first modeling sub-module is used to construct a corresponding three-dimensional space point set according to the first temperature value, the first humidity value, and the first conductivity value in the standard soil data, and respectively fit and generate three groups of correlation plane equations for the temperature-humidity plane, the electric-temperature plane, and the electric-humidity plane on the three-dimensional space point set;

[0074] The similarity calculation sub-module is used to map the standard soil data onto the correlation plane equation, calculate the normal distance between the mapped point and the correlation plane, and obtain three groups of spatial offset values;

[0075] The weight fusion sub-module is used to construct a temperature-humidity-conductivity correlation matrix according to the three groups of spatial offset values, perform normalization processing on it, extract the difference distribution characteristics between each parameter, and calculate the target matching coefficient.

[0076] In the embodiments of the present invention, in the multi-parameter soil detection task, there are often non-linear and time-varying implicit correlations among different soil properties. If directly judging whether the single-point values of temperature, humidity or conductivity are abnormal, it is very easy to cause false alarms or missed alarms. Therefore, in order to improve the overall perception ability of data anomalies, the system introduces a multi-dimensional modeling strategy through the correlation verification module to realize the joint judgment of the structural differences between parameters.

[0077] Among them, the first modeling sub-module constructs a three-dimensional space point set based on three types of parameters in the standard soil data, and respectively fits the temperature-humidity plane, the electric-temperature plane and the electric-humidity plane on this point set, so as to capture the joint change law between parameters. Compared with the traditional single linear fitting method, this strategy focuses more on the expression of the interactive geometric structure of the three types of parameters, which helps to explore the relative change trend and collaborative behavior between variables.

[0078] The similarity calculation sub-module maps the standard soil data of the current period to the above-mentioned fitting plane and calculates its normal distance, so as to obtain the spatial offset value of each plane. This geometric distance-based method can reflect the deviation degree of the current data on the fitting structure, and has more directionality and relevance than the traditional residual judgment.

[0079] The weight fusion sub-module further constructs a temperature-humidity-conductivity correlation matrix by unifying the three groups of spatial offset values, and based on the normalization and feature extraction operations, identifies the contribution weights of different parameters to the abnormal trend, so as to output the target matching coefficient. This matching coefficient is essentially a comprehensive quantification index of the data structure stability, with both dynamics and interpretability.

[0080] The combination of such correlation modeling and weight fusion not only enhances the tolerance of the system to minor perturbations, but also improves the detection ability of potential trend breaks and parameter mismatches, and has good adaptability and promotion value.

[0081] In a preferred embodiment of the present invention, the reconstruction module includes:

[0082] The time reversal sub-module is used to, after the abnormal prompt signal is activated, call the standard soil data in the preset detection points in the historical database within the preset number of recent non-abnormal detection cycles, and reverse the order of the data to generate a time-sequence reversal data set;

[0083] The trend prediction sub-module is used to construct a first-order state transition equation according to the historical values corresponding to the abnormal sub-items in the time-sequence reversal data set, and combine the known non-abnormal sub-items to perform trend fitting on the missing values by means of cross regression to generate a predicted interpolation;

[0084] The structural constraint submodule is used to introduce the upper tolerance constraint of the matching coefficient, perform error feedback correction on the predicted interpolation, eliminate the interpolation points that exceed the preset dynamic tolerance threshold range during the fitting process, and generate reconstructed soil data.

[0085] In the embodiment of the present invention, it is a common phenomenon that some parameters are missing or distorted due to instantaneous sensor failure, communication interruption or environmental interference in soil data collection. If effective reconstruction processing is not performed, it may directly affect the stability of the system output results. To this end, this system designs a reconstruction module based on historical co-location data combined with a time series fitting strategy to achieve intelligent replacement calculation of abnormal sub-items.

[0086] When the system identifies an abnormality at the current detection point through the matching coefficient, the time inversion submodule is activated, and the standard soil data of the same detection point with no abnormal records within the preset number of abnormal detection cycles are called from the historical database, and these data are arranged in reverse chronological order to form a time series inversion data set. This reverse order structure helps to capture the neighboring trend change characteristics of soil status over time.

[0087] The trend prediction submodule constructs a first-order state transfer equation based on the data set, models the evolution trend of abnormal sub-items in history, introduces non-abnormal sub-items as joint input variables, fits missing points through cross regression, and generates predictive interpolations, taking into account the correlation between variables and the laws of historical paths.

[0088] The structural constraint submodule further introduces the matching coefficient as the upper tolerance constraint factor, performs error feedback correction on the fitting results of the prediction interpolation, and eliminates the prediction points that exceed the dynamic tolerance threshold during the fitting process, ultimately generating reconstructed soil data with both structural consistency and data rationality.

[0089] This reconstruction path starts from the three dimensions of data integrity, variable structure retention, and historical consistency, ensuring that defect repair is not limited to numerical completion, but also emphasizes its environmental rationality and predictive reliability, significantly improving the system's ability to handle data loss or contamination situations, and showing good robustness and intelligence in practical applications such as continuous monitoring and multi-point deployment.

[0090] In a preferred embodiment of the present invention, the first modeling submodule includes:

[0091] A window screening unit is used to extract multiple data nodes within a preset time window according to standard soil data to form a three-dimensional space point set for modeling;

[0092] A plane fitting unit is used to fit the three-dimensional space point set and generate three sets of associated plane equations of the temperature-humidity plane, the electric-temperature plane and the electric-humidity plane;

[0093] The reference plane calibration unit is used to dynamically determine the initial normal vector direction of each group of planes according to the historical point set distribution statistics of different detection points.

[0094] In the embodiment of the present invention, during the soil detection process, the quality of the standard soil data directly affects the accuracy of subsequent anomaly identification, and whether the parameters have a stable internal structural relationship is the key to identifying abnormal change trends. Therefore, in the multidimensional modeling stage, in order to enhance the adaptive ability of plane fitting and improve its physical association significance, the system introduces structures such as window screening, plane fitting and benchmark calibration to achieve functional refinement of the first modeling submodule.

[0095] The window screening unit extracts a set of data nodes with temporal continuity from the standard soil data according to the preset time window to form a three-dimensional spatial point set. Through this strategy, only representative recent data is retained to avoid the interference of long-distance historical data on the modeling results, thus improving the timeliness and local effectiveness of the modeling results.

[0096] The plane fitting unit uses the extracted point set to construct the two-dimensional mapping relationship between the three groups of parameters, generate the temperature-humidity plane, the electric-temperature plane and the electric-humidity plane, and complete the expression of the local coupling relationship between soil parameters. This correlation plane based on the fitting of the three-dimensional point set reflects the co-evolution structure of the three types of parameters within a time window.

[0097] The reference plane calibration unit further combines the historical point set distribution characteristics of different detection points to dynamically correct the plane normal vector direction. By introducing this calibration mechanism, the fitting deviation problem that occurs under extreme climate or heterogeneous terrain conditions can be effectively avoided, thereby improving the practical adaptability of the plane model.

[0098] On the whole, through the structural refinement of the first modeling sub-module, the system achieves dynamic adaptation of the spatial correlation modeling process and stable enhancement of the fitting results without increasing the computational complexity, laying a high-quality foundation for subsequent offset judgment and matching coefficient calculation.

[0099] Among them, the role of the plane fitting unit is to establish a corresponding spatial fitting structure based on the relationship between different parameters in the standard soil data to reflect the linkage trend between temperature, humidity and conductivity under normal conditions. To this end, the system selects multiple data nodes that are continuous in time to form a three-dimensional spatial point set, and each point is composed of soil parameter values in three dimensions.

[0100] After constructing the point set, the system performs fitting operations respectively for three groups of variable combinations: temperature - humidity, conductivity - temperature, and conductivity - humidity. The specific fitting method adopts the minimum deviation criterion, that is, to find a geometric plane in three - dimensional space that can be closest to all data points. Through this fitting operation, the system generates three plane models, which respectively represent the conventional co - variation paths among the three types of parameters.

[0101] The fitting plane is not used for prediction, but as a reference structure for subsequent offset calculation and anomaly judgment. Once the data of a new detection cycle is input, the system can quickly determine whether it deviates from these planes, thereby assisting in identifying whether there are potential anomalies.

[0102] The advantage of this unit is that it avoids the complex formula modeling and training process, extracts the structural relationships among soil parameters through a data - driven approach, and has the characteristics of strong real - time performance and high adaptability. It is especially suitable for embedded environments that require low computational load.

[0103] Among them, the reference plane calibration unit is used to dynamically correct the directionality of the fitting model. In the actual environment, due to regional differences in soil composition, temperature distribution, and humidity gradient, directly using a unified fitting method for data from different locations may cause the direction of the fitting plane to deviate from the main trend of the data.

[0104] Therefore, the system conducts distribution statistics on the point sets generated within the historical cycle of each detection point, calculates its main change direction. The reference plane calibration unit uses this change direction as a reference to perform normal vector rotation or offset correction on the original fitting plane, making it closer to the true evolution trend of local soil data. This process is achieved through data distribution weight adjustment, does not rely on specific calculation formulas, and has good versatility.

[0105] The calibration result is stored as the initialization parameters of the fitting model for the detection point and is directly called during subsequent fitting, ensuring that even in cases such as system restart and device replacement, it can quickly return to the adapted state, improving the system deployment efficiency and long - term operation stability.

[0106] Through the introduction of this unit, while maintaining the lightweight overall architecture, the adaptability to different soil environments can be significantly improved, ensuring that the fitting model always has reliability and practical significance in the long term.

[0107] In a preferred embodiment of the present invention, the similarity calculation sub - module includes:

[0108] A projection transformation unit, which is used to map the standard soil data in the current detection cycle to spatial detection points, project them onto the corresponding plane in the associated plane equation, and obtain the projected point data;

[0109] A distance calculation unit, configured to calculate a temperature-humidity offset value, an electric-temperature offset value, and an electric-humidity offset value respectively according to the normal vertical distance between the projection point and the original space detection point, and form a set of space offset values;

[0110] An offset adjustment unit, configured to perform an anomaly marking on the offset values in the set of space offset values whose fluctuations exceed a preset stability threshold.

[0111] In an embodiment of the present invention, in the anomaly recognition link of soil data, the absolute numerical values between parameters often cannot fully reflect the potential anomalies of the system. The collaborative relationship and trend offset between parameters are more valuable judgment bases. Therefore, after the standard soil data is modeled, the system introduces a similarity calculation sub-module, which is used to accurately quantify the offset degree between the current detection period and the historical correlation structure.

[0112] In the actual execution process, the projection transformation unit first converts the standard soil data in the current detection period into space detection points and projects them onto the pre-constructed correlation plane equation. This operation not only retains the geometric attributes of the original data but also explicitly establishes the mapping relationship between the current data and the known stable structure.

[0113] Based on the normal vertical distance between the projection point and the original space detection point, the distance calculation unit calculates a temperature-humidity offset value, an electric-temperature offset value, and an electric-humidity offset value respectively. These offset values, as the quantization indexes of the deviation of the space structure, have directionality and physical interpretability and can be used as the core basis for judging the fluctuations of the soil environment.

[0114] The offset adjustment unit further analyzes the above offset values based on the historical stability threshold and marks the items with abnormal fluctuations. This processing step not only provides a data basis for the subsequent weight fusion operation but also provides an early warning mechanism for anomaly recognition and trend monitoring.

[0115] Through the mapping and offset mechanism of this module, the system has the ability to identify structural mutations of parameters and judge the overall deviation trend of data, thereby improving the recognition accuracy of atypical anomaly patterns and expanding the applicable range and intelligent perception depth of the soil detection system.

[0116] Among them, in order to realize the correlation judgment between the standard soil data and the fitting plane, the core function of the projection transformation unit is to convert the three-dimensional parameter data (i.e., the first temperature value, the first humidity value, and the first conductivity value) in the current detection period into space points and project the points onto the correlation plane generated by the aforementioned fitting. In the specific operation process, the system constructs the current soil data in the form of three-dimensional coordinates and constructs projection points in the sense of space geometry according to the structural parameters of each fitting plane.

[0117] This projection operation is used to find the shortest mapping path of data points in the associated plane, so as to measure the degree of deviation of the data from the model structure. Since this unit only focuses on geometric position relationships and does not need to perform complex regression analysis, it has the advantages of simple operation and high deployment efficiency. It is particularly suitable for resource-constrained embedded processors or low-power sensing platforms.

[0118] Through this operation, the system can convert the original sampled data into structure-associated data, providing a clear spatial basis for subsequent offset value calculation and stability analysis.

[0119] Among them, the distance calculation unit is used to obtain the normal distance difference between the original detection point and its projection point to reflect the degree of offset of the soil parameters from the fitting plane. This unit processes three groups of structures, namely the temperature-humidity plane, the electric-temperature plane, and the electric-humidity plane, and outputs a corresponding offset value for each group.

[0120] This operation is processed using the vector difference or the shortest distance principle, and the result has dual meanings of directionality and magnitude. The system judges whether the current detection point remains within the established associated structure range through this distance value. If the offset is large, it indicates that there is an abnormal trend or a decrease in stability in the data of this dimension.

[0121] The implementation of this unit depends on the preposed geometric projection process. The overall logic is simple and it is easy to implement through standard matrix operations or table lookup methods. It has the characteristics of high generality and low computational burden in the case of multi-point parallel deployment.

[0122] Among them, the function of the offset adjustment unit is to analyze the three groups of calculated spatial offset values and identify the parts whose fluctuations exceed the preset stability threshold. In the actual environment, soil data will be affected by temperature and humidity fluctuations, sunlight radiation, or human intervention, resulting in instantaneous anomalies. Therefore, judging anomalies based on only one group of offset values may lead to misjudgments.

[0123] For this reason, this unit screens each offset value according to the historical statistical threshold, marks the abnormal offset items, and the marking information will be an important input when calculating the target matching coefficient subsequently. The processing method of this unit supports dynamic threshold update to adapt to the differences in soil properties in different regions.

[0124] Through this processing mechanism, the system has the dual capabilities of identifying sudden disturbances and trend anomalies, further improving the accuracy of soil data anomaly judgment and the fine-grained control level.

[0125] In a preferred embodiment of the present invention, the weight fusion sub-module includes:

[0126] A weight allocation unit, configured to allocate weighting coefficients for temperature-humidity offset values, electric-temperature offset values, and electric-humidity offset values according to the historical stability scores of sensors corresponding to preset detection points, and generate an initial offset matrix, where each column of the initial offset matrix corresponds to three sets of spatial offset values;

[0127] A difference extraction unit, configured to perform normalization processing on the initial offset matrix, construct a temperature-humidity correlation matrix, and extract the difference distribution characteristics between different parameter combinations in the matrix, where the difference distribution characteristics are calculated based on the covariance distance between parameters;

[0128] A matching calculation unit, configured to input the difference distribution characteristics into a preset weighted scoring model, and calculate a target matching coefficient in the current detection period.

[0129] In an embodiment of the present invention, in a scenario where there are multiple offset values between parameters, how to effectively evaluate the impact of these offsets on the overall stability of the detection point is the key to achieving accurate anomaly determination. Based on this requirement, the system structurally integrates the spatial offset values through a weight fusion sub-module, introduces a weighted scoring mechanism and a difference extraction process, and completes the generation of the target matching coefficient.

[0130] The weight allocation unit first assigns different weighting coefficients to the three types of offset values according to the historical stability scores of the sensors used for the preset detection points, and generates an initial offset matrix. This weight not only reflects the stability degree of different sensor dimensions, but also can dynamically evaluate the signal source at the system level, so that the overall matching has stronger context relevance.

[0131] The difference extraction unit performs normalization processing on the initial offset matrix, constructs a temperature-humidity correlation matrix, and extracts the collaborative offset structure between various parameters based on indexes such as covariance distance. Through this process, the system not only identifies the anomalies of single data, but also discovers the structural fluctuations caused by the coupling imbalance between parameters.

[0132] The matching calculation unit uses the extracted difference distribution characteristics, inputs them into a preset weighted scoring model, and outputs the target matching coefficient in the current detection period. This matching coefficient, as the core index for subsequent anomaly judgment and reconstruction processing, has the advantages of real-time calculation, strong interpretability, and flexible adjustment.

[0133] Through the scoring mechanism design of the weight fusion sub-module, the system can adaptively identify high-impact factor parameters, quantitatively model and evaluate their contribution degrees to the overall data stability, provide highly reliable data basis for subsequent modules, and effectively improve the reliability and adaptability of the soil detection system in a complex changing environment.

[0134] Among them, the weight allocation unit assigns weighting coefficients based on the historical stability scores of each sensor. The stability scores can be comprehensively scored according to dimensions such as the historical working state of the device, the error frequency, and the aging condition of the sensor, ensuring a perceptual basis for the data fusion process.

[0135] This scoring is usually analyzed once during system initialization or regular operation, and then the weight table is updated for the current detection cycle. After the allocation is completed, the three groups of spatial offset values are stored in the initial offset matrix according to their respective weights, providing a fusion basis for the difference extraction unit.

[0136] Through this allocation method, the system can establish an effective trust ranking mechanism among multi-source heterogeneous signals, making the matching judgment results more in line with the actual physical state, and helping to reduce the judgment error caused by a single outlier.

[0137] Among them, the difference extraction unit takes the initial offset matrix as the basis and normalizes it so that each parameter participates in the analysis under the same dimension. Then, the system constructs a temperature-humidity-conductivity correlation matrix to describe the joint fluctuation characteristics between each offset item. Specifically, the system analyzes the mutual relationship of the offset degrees between different parameter combinations (such as temperature-humidity, humidity-conductivity, etc.) to identify whether there is a structural mismatch.

[0138] This process does not rely on a fixed model, but dynamically constructs the matrix structure according to the distribution characteristics of the actual detection data. Therefore, the system can automatically adapt and adjust the analysis model according to different soil types and monitoring environments, and has strong generalization ability.

[0139] Through the difference analysis of this subunit, the system not only retains the independent fluctuation information of each parameter, but also extracts the coupling relationship between them, providing a structural level support basis for the final matching degree calculation.

[0140] Among them, the matching calculation unit completes the calculation of the target matching coefficient according to the parameter distribution characteristic information output by the difference extraction unit and in combination with the configured weighted scoring model. The matching coefficient, as the core index to measure the structural stability of the soil state in the current detection cycle, will be used in anomaly judgment and trend repair.

[0141] The scoring model can adopt a multi-factor weighting method, combining factors such as the size of the offset value, historical stability, and fluctuation trend to form a comprehensive judgment result. This calculation process provides a judgment basis for the subsequent module. If its output value is lower than the set stability threshold, an anomaly prompt process will be triggered.

[0142] The introduction of this subunit enables the system to have the ability to quantitatively evaluate the stable state of multi-dimensional soil data, transforming from the "single value over-limit" judgment mode to the "structural deviation" comprehensive analysis mode, effectively improving the data perception depth and the system operation stability.

[0143] In a preferred embodiment of the present invention, the time reversal sub-module includes:

[0144] A historical screening unit, configured to screen data segments in the historical database that have a similarity higher than a similarity threshold with the environmental parameters of the current detection period within the last three anomaly-free detection periods, and generate an inversion reference data source;

[0145] A time series sorting unit, configured to sort the standard soil data in the inversion reference data source in reverse order according to the acquisition timestamp, and maintain the corresponding relationships of the first temperature value, the first humidity value, and the first conductivity value, so as to construct a preliminary time series inversion data set;

[0146] A boundary cleaning unit, configured to perform extreme value identification and elimination operations on the boundary points in the preliminary time series inversion data set, and generate a time series inversion data set.

[0147] In the embodiments of the present invention, before performing the abnormal sub-item substitution calculation, selecting what historical data as the reference basis is a decisive factor affecting the subsequent prediction quality and fitting accuracy. To improve the pertinence and rationality of time series construction, the system sets up a time reversal sub-module, which significantly enhances the adaptability of historical data in the fitting stage by introducing similarity screening and time series structure reconstruction.

[0148] First of all, the historical screening unit calculates the similarity between the environmental parameters of the current detection period and the historical period, and sets a similarity threshold, and only screens out the data segments of the detection periods that are similar to the current conditions and have no anomalies to form an inversion reference data source. Compared with the traditional sliding window historical callback, this method can avoid the problem of inconsistent structures caused by external conditions such as climate mutations and soil layer changes, thus ensuring the reference value of historical data.

[0149] The time series sorting unit performs reverse sorting on the reference data based on the acquisition timestamp, so that the data closest to the current time is arranged in the front position. During the process of reconstructing the data set, the corresponding relationships between the first temperature value, the first humidity value, and the first conductivity value are kept unchanged, effectively retaining the time series integrity of the data structure and providing a prior basis for constructing a high-quality first-order state transition equation.

[0150] The boundary cleaning unit performs extreme value identification and elimination operations on the extreme points at the front and rear ends of the data set. Especially in the case of abnormal jumps, it can avoid the fitting deviation caused by short-term mutation values to the subsequent model construction, and enhance the smoothness and trend consistency of the time series data.

[0151] The time series inversion data set constructed through the above three steps has the characteristics of strong time effectiveness correlation, consistent environmental conditions, and low noise interference. It not only improves the prediction accuracy of the model, but also enhances the repair reliability of the system for abnormal samples, providing a data basis with controllable structure for the entire reconstruction path.

[0152] Among them, during the process of constructing the time series inversion dataset, since the historical data may contain instantaneous jumps, sensor fluctuations, or short-term abnormal data, these data are usually located at the edges of the time series and are extremely likely to interfere with the stability of subsequent trend modeling. To improve the reliability and smoothness of the inversion dataset, the system introduces a boundary cleaning unit to perform extreme value identification and elimination operations on the boundary points in the preliminarily constructed time series inversion dataset.

[0153] In the specific implementation process, the boundary cleaning unit uses a rolling window method to scan a number of time nodes before and after in the time series inversion dataset, and extracts the corresponding first temperature value, first humidity value, and first conductivity value. For each type of parameter, the system conducts statistical analysis on its historical value range to determine whether the point significantly deviates from the historical mean or fluctuation range. If a data point exceeds the set upper and lower bound tolerance range, it is marked as a boundary extreme value point.

[0154] The identified boundary extreme value points will be automatically eliminated, or in extremely rare cases, data smoothing processing (such as median substitution) will be performed, so as to ensure that the inversion dataset has sufficient time consistency and trend continuity before constructing the state transition model. This mechanism is especially applicable to scenarios with frequent sudden disturbances such as horticultural greenhouses, and improves the quality of the basic data for trend modeling.

[0155] In a preferred embodiment of the present invention, the trend prediction sub-module includes:

[0156] A state modeling unit, which is used to construct a first-order linear state transition equation according to the historical values corresponding to the abnormal sub-items in the time series inversion dataset and introduce a time-decreasing weight to apply a decay factor to the long-distance historical data;

[0157] A cross-fitting unit, which is used to introduce the real-time data of the non-abnormal sub-items as joint input variables into the first-order linear state transition equation for cross-regression modeling, and generate a preliminary prediction interpolation while maintaining the correlation of the existing variables;

[0158] A confidence evaluation unit, which is used to calculate the sum of squared fitting residuals of the preliminary prediction interpolation and compare it with a preset confidence threshold. If the sum of squared residuals is lower than the confidence threshold, the corresponding preliminary prediction interpolation is marked as a valid prediction value to generate a prediction interpolation.

[0159] In the embodiment of the present invention, after the construction of the inversion dataset is completed, how to perform trend prediction on the abnormal sub-items is a key issue to ensure the quality of soil data reconstruction. The trend prediction sub-module designs a state modeling unit, a cross-fitting unit, and a confidence evaluation unit around two core logics of historical trend modeling and variable cross-fitting, ensuring that the prediction results have sufficient fitting accuracy and determination reliability.

[0160] Based on the historical values of abnormal sub-items in the time-series inversion dataset, the state modeling unit constructs a first-order linear state transition equation and introduces a time-decreasing weight, setting the weight of data closer to the current period higher and the weight of distant data gradually decaying. This time decay mechanism can better approximate the short-period variation characteristics of soil data and enhance the representativeness of trend modeling in the current period.

[0161] The cross-fitting unit takes the real-time data of non-abnormal sub-items in the current period as joint input variables into the state transition equation and performs cross-regression modeling operations. On the premise that the correlation between variables remains unchanged, this strategy can use the data of non-abnormal sub-items as an "inference intermediary" to indirectly repair the missing trends of abnormal sub-items, which is particularly applicable to scenarios where there are coupling relationships between soil parameters.

[0162] The confidence evaluation unit calculates the sum of squared fitting residuals for the preliminary predicted interpolation and compares it with a preset confidence threshold to eliminate low-confidence results and screen out predicted interpolations that are closer to the historical evolution path. This result verification mechanism can effectively prevent misrepair phenomena caused by data fluctuations or model overfitting.

[0163] Overall, the trend prediction sub-module realizes a closed-loop logic from time modeling to variable association and then to reliable output, providing a prediction mechanism with both structural interpretability and output reliability for the system when facing uncertain environments and non-linear change trends.

[0164] In a preferred embodiment of the present invention, the structural constraint sub-module includes:

[0165] An error evaluation unit, which is used to calculate a matching error score according to the difference between the predicted interpolation and the standard soil data, in combination with the sum of squared fitting residuals of the predicted interpolation; where,

[0166] ,

[0167] is the matching error score, , , are weight coefficients, satisfying , is the current predicted interpolation, is the value of the corresponding sub-item in the standard soil data, is the predicted value of the th sub-item, such as temperature, humidity, conductivity, is the original standard value of the th sub-item, is the number of sub-items, is the target matching coefficient for the current detection period;

[0168] A tolerance limit unit for setting a dynamic tolerance threshold based on a matching coefficient, comparing the matching error score with the dynamic tolerance threshold, and screening out valid interpolation points within the range of the dynamic tolerance threshold;

[0169] A feedback correction unit for performing quadratic fitting processing based on the principle of residual minimization on interpolation points that do not meet the range of the dynamic tolerance threshold, and merging the corrected interpolation points with the valid interpolation points to generate reconstructed soil data.

[0170] In the embodiment of the present invention, even if high-confidence interpolation values are constructed through trend prediction, in the actual soil application scenario, the data itself is vulnerable to micro-environment fluctuations, and single residual control is difficult to meet the accuracy requirements. To further improve the structural consistency and prediction rationality of the interpolation data, a structure constraint sub-module is designed in the system to achieve multi-level error evaluation and tolerance feedback processing.

[0171] The error evaluation unit first calculates the matching error score according to the difference between the predicted interpolation and the standard soil data, combined with the sum of squared fitting residuals of the predicted interpolation. This score integrates the dual information of prediction deviation and model fitting residuals, and more comprehensively reflects the credibility of the interpolation.

[0172] The tolerance limit unit sets a dynamic tolerance threshold based on the matching coefficient calculated in the current cycle, and compares and screens it with the matching error score. This method introduces dynamic upper and lower limit control capabilities, and can automatically adjust the error tolerance boundary according to the complexity of the soil environment, thereby improving the intelligent judgment ability of the system.

[0173] The feedback correction unit performs quadratic fitting processing based on the principle of residual minimization for interpolation points that do not meet the tolerance requirements. If the tolerance condition still cannot be met after multiple fittings, the interpolation point is excluded. Finally, only the interpolation that meets the structural control conditions is merged with the original valid data to form complete and high-confidence reconstructed soil data.

[0174] This structure constraint process not only improves the rationality of the interpolation by introducing the judgment of structural consistency between data and the optimization path of residual minimization, but also enhances the elastic control ability of the system for non-linear anomaly compensation, showing strong adaptability and robustness in the actual deployment scenario.

[0175] Among them, a sub-item refers to a single physical parameter dimension in the standard soil data, that is:

[0176] The first temperature value (representing soil temperature), the first humidity value (representing soil water content), and the first conductivity value (representing soil salinity or nutrient ion concentration);

[0177] These three types of parameters appear as the coordinate dimensions of three-dimensional space points in the system, and each dimension corresponds to a sub-item.

[0178] Among them, the feedback correction unit undertakes the secondary optimization task of abnormal interpolation points, and its goal is to improve the credibility of the reconstructed soil data and reduce the misjudgment caused by the preliminary fitting error.

[0179] During the implementation process, when the predicted interpolation is identified by the tolerance limit unit as not meeting the dynamic tolerance threshold, the system will hand over these abnormal interpolation points to the feedback correction unit for processing. This unit reconstructs the local fitting environment by calling the historical data of adjacent time points in the time series inversion dataset, and performs a secondary fitting operation based on the principle of residual minimization. Compared with the preliminary interpolation, this operation pays more attention to the local data structure, and finely adjusts the interpolation result by analyzing the sample density and trend direction around the predicted value.

[0180] If an interpolation point still cannot reach the error tolerance range after more than one round of correction, the system will mark this point as untrusted data and exclude it from the reconstruction process to prevent it from interfering with the overall stability of the final data output.

[0181] By introducing the feedback correction unit, the system establishes a closed-loop reconstruction mechanism with error correction ability, making the interpolation calculation no longer a one-way operation, but a dynamic process that can be rolled back and optimized, significantly enhancing the self-adaptability in a changing environment.

[0182] In a preferred embodiment of the present invention, the cross-fitting unit includes:

[0183] A variable scoring sub-unit, which is used to extract the historical values of abnormal sub-items from the time series inversion dataset based on the first-order linear state transition equation, and use the real-time data of non-abnormal sub-items as joint input variables to calculate the correlation coefficient and its predicted residual variance between each non-abnormal sub-item and the abnormal sub-item, and generate an influence value; where

[0184] ,

[0185] is the variable influence degree of the th non-abnormal sub-item, , , are weight coefficients, satisfying , is the actual value of the th non-abnormal sub-item in the th historical detection period, is the th historical average value of the non-abnormal sub-item, is the actual value of the abnormal sub-item in the th historical detection period, is the historical average value of the abnormal sub-item. is the total number of historical detection cycles, is the th non - abnormal sub - item's predicted value in the th detection cycle, is the th sub - item's matching coefficient within the th detection cycle, is the target matching coefficient of the current detection cycle;

[0186] The model selection sub - unit is used to judge the prediction effectiveness of non - abnormal sub - items according to the influence degree value. If the influence degree value is higher than the preset influence degree threshold, the non - abnormal sub - item is used as an input variable to construct a multiple - linear state - transfer path; if the influence degree value is lower than the preset influence degree threshold, it is input into the regression path constructed based on the support - vector structure to generate two groups of candidate prediction interpolations respectively;

[0187] The prediction combination sub - unit is used to perform weighted fusion on the two groups of prediction interpolations to generate a preliminary prediction interpolation.

[0188] In the embodiment of the present invention, during the cross - regression process, whether the non - abnormal sub - item has prediction value will directly affect the stability and accuracy of the reconstruction result. Therefore, the system embeds a cross - fitting sub - module in the trend prediction sub - module to refine the modeling process, and realizes the intelligent switching of the model path and prediction integration through three sub - structures: variable scoring, model selection, and prediction combination.

[0189] The variable scoring sub - unit extracts the historical values of abnormal sub - items based on the first - order state - transfer equation, uses the non - abnormal sub - items of the current cycle as input variables, calculates the correlation coefficient and prediction residual variance between each non - abnormal sub - item and the abnormal sub - item, and then quantifies them into influence degree values. This scoring mechanism takes into account both the synergy and prediction stability between variables to ensure that the selected variables have actual prediction ability.

[0190] The model selection sub - unit takes the influence degree value as a criterion. If the influence degree of the non - abnormal sub - item is higher than the preset threshold, it is included in the multiple - linear state - transfer path for prediction; if it is lower than the threshold, a non - linear regression path is constructed using the support - vector structure. This design makes the model selection process data - driven, independent of preset preferences, and effectively improves the adaptability under different data states.

[0191] The prediction combination sub - unit performs weighted fusion on the candidate prediction interpolations output by the two regression models according to the historical stability and the prediction confidence of the current cycle, and outputs a comprehensive prediction result. This kind of weighted mechanism not only integrates the advantages of multiple models but also retains the weight difference of each prediction path, making the final interpolation more smooth and controllable.

[0192] Through the introduction of this sub-module, the entire trend prediction path has transitioned from a single inference mechanism to a dynamic modeling mechanism based on scoring feedback, significantly enhancing the adaptability elasticity and result control ability of the prediction model, and improving the overall response accuracy of the system to soil data fluctuations.

[0193] Among them, before performing cross-regression modeling, the system needs to determine which non-abnormal sub-items have sufficient predictive ability and can be used as inference variables for abnormal sub-items. For this purpose, the variable scoring sub-unit undertakes the core screening function during the cross-fitting process.

[0194] This unit extracts the values of non-abnormal sub-items in multiple historical periods from the time-series inversion dataset and pairs them with the historical values of abnormal sub-items in the corresponding periods for analysis. By statistically analyzing the consistency of the change trends and the degree of deviation between each pair of variables, the system can calculate the predictive contribution degree of each non-abnormal sub-item to the abnormal sub-item, forming an influence value.

[0195] The level of influence is not only affected by the correlation between variables but also related to the fluctuation range of historical prediction deviations. This sub-unit integrates and evaluates multiple factors, screening out those non-abnormal sub-items with stable predictive ability and consistent trend directions in historical data as the input for subsequent model construction.

[0196] This scoring mechanism improves the quality of variable selection in the regression model, avoids the fitting interference caused by redundant variables, and is conducive to enhancing the credibility and interpretability of predictive interpolation.

[0197] Among them, after completing the influence analysis, the system does not use a unified modeling method to perform predictions on all variables, but realizes a differential path design through the model selection sub-unit.

[0198] For non-abnormal sub-items with high influence values, the system selects to construct a multi-linear state transition path to predict the trend of abnormal sub-items through the existing state modeling logic; for variables with low influence values, because their predictive ability is greatly interfered by noise, the system uses regression models with non-linear fitting abilities such as support vector regression (SVR) for processing.

[0199] The key of this sub-unit lies in dynamically judging the modeling mechanism applicable to each type of variable, so that different types of variables can play a predictive role in their most suitable paths. Through the introduction of the path selection mechanism, the system introduces structural elasticity in the modeling process, effectively avoiding the misjudgment problem of "model rigidity" for complex data structures.

[0200] Among them, after multiple regression models respectively output predictive interpolation, the system needs to perform fusion to generate a final predictive result with stronger stability. The prediction combination sub-unit is responsible for integrating the output results of each regression path and performing weighted processing according to multiple dimensions.

[0201] The setting of the weighting coefficient is based on two core factors: one is the prediction stability of the variable in the historical period, which is manifested as the variance value of the prediction residual; the other is the confidence score in the current period, which reflects the reliability of the model under the current input. The system generates a weighting factor through these two indicators, performs weighted average processing on each candidate prediction interpolation, and outputs a preliminary prediction interpolation.

[0202] This sub-unit not only improves the smoothness and consistency of the final result, but also effectively integrates the advantages of different models, which is a key step in realizing integrated optimization in the entire cross-fitting mechanism.

[0203] The above is the preferred embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art of the present technology, without departing from the principle described in the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A multi-functional soil detection system, characterized in that, The system comprises: The correlation verification module is used to construct a temperature-humidity-conductivity correlation matrix according to the standard soil data, and obtain a set of matching coefficients by verifying the difference distribution between the first temperature value and the first humidity value and the first conductivity value in the matrix; An abnormality judgment module is used to judge whether there is a data reading abnormality at a preset detection point according to the matching coefficient, and generate an abnormality prompt signal if the matching coefficient is lower than a preset stability threshold; The reconstruction module is used to perform alternative calculations on the abnormal sub-items in the standard soil data according to the abnormal prompt signal, and generate reconstructed soil data by curve fitting based on historical co-site data; An integrated output module is used to merge the reconstructed soil data with the standard soil data without abnormalities to obtain the final soil detection data; The association verification module comprises: The first modeling submodule is used to construct a corresponding three-dimensional space point set according to the first temperature value, the first humidity value and the first conductivity value in the standard soil data, and to fit and generate three sets of associated plane equations of the temperature-humidity plane, the electric-temperature plane and the electric-humidity plane on the three-dimensional space point set respectively; The similarity calculation submodule is used to map the standard soil data onto the associated plane equation, calculate the normal distance between the mapping point and the associated plane, and obtain three sets of spatial offset values; The weight fusion submodule is used to construct the temperature and humidity correlation matrix based on the three sets of spatial offset values, normalize it, extract the difference distribution characteristics between the parameters, and calculate the target matching coefficient.

2. The multifunctional soil detection system according to claim 1, characterized in that The reconstruction module includes: The time inversion submodule is used to call the standard soil data within the most recent preset number of abnormal detection cycles of the preset detection points in the historical database after the abnormal prompt signal is activated, and arrange the data in reverse order to generate a time series inversion data set; The trend prediction submodule is used to construct a first-order state transfer equation based on the historical values corresponding to the abnormal sub-items in the time series inversion data set, combine the known non-abnormal sub-items, use the cross regression method to perform trend fitting on the missing values, and generate prediction interpolation; The structural constraint submodule is used to introduce the upper tolerance constraint of the matching coefficient, perform error feedback correction on the predicted interpolation, eliminate the interpolation points that exceed the preset dynamic tolerance threshold range during the fitting process, and generate reconstructed soil data.

3. The multifunctional soil detection system according to claim 1, characterized in that, The first modeling submodule includes: A window screening unit is used to extract multiple data nodes within a preset time window according to standard soil data to form a three-dimensional space point set for modeling; A plane fitting unit is used to fit the three-dimensional space point set and generate three sets of associated plane equations of the temperature-humidity plane, the electric-temperature plane and the electric-humidity plane; The reference plane calibration unit is used to dynamically determine the initial normal vector direction of each group of planes according to the historical point set distribution statistics of different detection points.

4. A multifunctional soil detection system according to claim 3, characterized in that, The similarity calculation submodule includes: A projection transformation unit is used to map the standard soil data in the current detection cycle into a spatial detection point, project it onto the corresponding plane in the associated plane equation, and obtain projection point data; A distance calculation unit, used to calculate the temperature and humidity offset value, the electrical temperature offset value and the electrical humidity offset value according to the normal vertical distance between the projection point and the original space detection point, to form a space offset value set; An offset adjustment unit for abnormally marking the offset values in the spatial offset value set that fluctuate beyond a preset stability threshold.

5. A multifunctional soil detection system according to claim 4, characterized in that The weight fusion sub-module includes: A weight assignment unit for assigning weighting coefficients to the temperature-humidity offset value, the electric temperature offset value, and the electric humidity offset value according to the historical stability scores of the sensors corresponding to the preset detection points, generating an initial offset matrix, where each column of the initial offset matrix corresponds to three groups of spatial offset values; A difference extraction unit for normalizing the initial offset matrix, constructing a temperature-humidity correlation matrix, and extracting the difference distribution characteristics between different parameter combinations in the matrix, where the difference distribution characteristics are calculated based on the covariance distance between parameters; A matching calculation unit for inputting the difference distribution characteristics into a preset weighted scoring model to calculate the target matching coefficient in the current detection period.

6. The multifunctional soil detection system according to claim 2, wherein, The time reversal sub-module includes: A historical screening unit for screening data segments in the historical database that have a similarity higher than the similarity threshold with the environmental parameters of the current detection period in the last three anomaly-free detection periods, generating an inversion reference data source; A time sequence sorting unit for sorting the standard soil data in the inversion reference data source in reverse order according to the acquisition timestamps and maintaining the corresponding relationships of the first temperature value, the first humidity value, and the first conductivity value to construct a preliminary time sequence inversion data set; A boundary cleaning unit for performing extreme value identification and elimination operations on the boundary points in the preliminary time sequence inversion data set to generate a time sequence inversion data set.

7. A multifunctional soil detection system according to claim 6, characterized in that, The trend prediction sub-module includes: A state modeling unit for constructing a first-order linear state transition equation according to the historical values corresponding to the abnormal sub-items in the time sequence inversion data set and introducing a time-decreasing weight to apply an attenuation factor to the long-distance historical data; A cross-fitting unit for introducing the real-time data of the non-abnormal sub-items as joint input variables into the first-order linear state transition equation for cross-regression modeling to generate a preliminary prediction interpolation while maintaining the correlation of the existing variables; A confidence evaluation unit for calculating the sum of squared fitting residuals of the preliminary prediction interpolation and comparing it with a preset confidence threshold. If the sum of squared residuals is lower than the confidence threshold, the corresponding preliminary prediction interpolation is marked as a valid prediction value to generate a prediction interpolation.

8. A multifunctional soil detection system according to claim 7, characterized in that, The structure constraint sub-module includes: An error evaluation unit for calculating a matching error score according to the difference between the prediction interpolation and the standard soil data, combined with the sum of squared fitting residuals of the prediction interpolation; A tolerance limit unit for setting a dynamic tolerance threshold based on the matching coefficient and comparing the matching error score with the dynamic tolerance threshold to screen out the valid interpolation points within the range of the dynamic tolerance threshold; A feedback correction unit for performing a quadratic fitting process on the interpolation points that do not meet the range of the dynamic tolerance threshold based on the principle of residual minimization and merging the corrected interpolation points with the valid interpolation points to generate reconstructed soil data.

9. A multifunctional soil detection system according to claim 8, characterized in that, The cross-fitting unit includes: A variable scoring sub-unit, which is used to extract the historical values of abnormal sub-items from the time series inversion dataset based on the first-order linear state transition equation, and use the real-time data of non-abnormal sub-items as joint input variables to calculate the correlation coefficient and its predicted residual variance between each non-abnormal sub-item and the abnormal sub-item, and generate an influence value; A model selection sub-unit, which is used to judge the prediction effectiveness of non-abnormal sub-items according to the influence value. If the influence value is higher than the preset influence threshold, the non-abnormal sub-item is used as an input variable to construct a multi-linear state transition path; if the influence value is lower than the preset influence threshold, it is input into the regression path constructed based on the support vector structure to generate two groups of candidate prediction interpolations respectively; A prediction combination sub-unit, which is used to perform weighted fusion on the two groups of prediction interpolations to generate a preliminary prediction interpolation.

Images