A method for detecting the moisture content of high-moisture-content soil based on smart sensors

Abstract

This invention discloses a method for detecting the moisture content of high-moisture-content soil based on intelligent sensors, belonging to the field of soil testing technology. The method includes: establishing spatiotemporal stratigraphic identification information in the high-moisture-content soil area based on intelligent sensors, collecting operational context parameters, and forming deployment context data; performing short-term trial sampling based on the deployment context data, screening candidate measurement mechanisms and extracting response offset feature groups, competitively scoring and dynamically ranking the candidate measurement mechanisms to generate an initial acquisition strategy; performing data alignment and confidence weighting across sensors, depth layers, and time windows based on calibration observation sequences, and outputting moisture content estimates, estimation confidence levels, and environmental sensitivity indicators through a preset moisture content inference mechanism to form judgment data. By iteratively generating calibration priors through virtual working conditions, the method improves the adaptability and calibration accuracy of the detection process in high-moisture-content soil scenarios.

Description

Technical Field

[0001] This invention relates to the field of soil testing technology, and in particular to a method for detecting the moisture content of high-moisture-content soil based on intelligent sensors. Background Technology

[0002] In recent years, intelligent sensors have continued to develop in the field of soil moisture content detection. Related technologies have gradually evolved from single-point static measurement to dynamic detection methods that combine multi-parameter sensing, time-series sampling, and model inference. These methods can collaboratively acquire multi-dimensional information such as regional identification, stratigraphic information, ambient temperature, conductivity, and sensor status, and form a foundation for continuous application in complex scenarios in agricultural monitoring, geological exploration, and ecological environment assessment.

[0003] Under high moisture content soil conditions, the sensing response is easily affected by electrical interference, local disturbances, response saturation and time drift. Simply relying on fixed sampling methods or static correction relationships makes it difficult to synchronize the selection of measurement mechanisms and correction criteria with the current working conditions, which in turn restricts the availability of the observation sequence and the reliability of subsequent moisture content inference. Therefore, it is urgent to improve the adaptive correction capability of the detection process to complex working conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for detecting the moisture content of high-moisture-content soil based on intelligent sensors to solve the problem of insufficient adaptive correction capability for moisture content detection under complex working conditions of high-moisture-content soil.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for detecting the moisture content of high-moisture-content soil based on intelligent sensors. The method includes: establishing spatiotemporal stratigraphic identification information in the high-moisture-content soil area using intelligent sensors, collecting operational context parameters, and forming deployment context data; performing short-term trial sampling based on the deployment context data, screening candidate measurement mechanisms and extracting response offset feature groups, competitively scoring and dynamically ranking the candidate measurement mechanisms, and generating an initial acquisition strategy; constructing a virtual operating condition state based on the initial acquisition strategy, response offset feature groups, and deployment context data, performing virtual operating condition iteration, obtaining correction priors, and forming a virtual correction strategy package; performing formal real-time sampling according to the virtual correction strategy package, obtaining the original observation sequence, inputting the original observation sequence and correction priors into a response quality decision model, outputting correction parameters, and performing correction processing on the original observation sequence based on the correction parameters to form a calibrated observation sequence; performing data alignment and confidence weighting across sensors, depth layers, and time windows based on the calibrated observation sequence, and outputting moisture content estimates, estimated confidence levels, and environmental sensitivity indicators through a preset moisture content inference mechanism to form judgment data.

[0007] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the spatiotemporal stratigraphic identification information includes regional identifiers, node identifiers, spatial location identifiers, depth stratigraphic identifiers, and time identifiers.

[0008] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific steps for forming deployment context data are as follows: Based on the corresponding region identifier, node identifier, spatial location identifier, depth layer identifier, and time identifier, perform integrity verification and consistency matching to obtain spatiotemporal layer identifier information; The runtime context parameters are aligned to the execution time, normalized to the fields, and screened for outliers. They are then associated and encapsulated with the corresponding spatiotemporal layer identifiers to form deployment context data.

[0009] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific steps for screening candidate measurement mechanisms and extracting response offset feature groups are as follows: Based on the environmental temperature parameters, electrical conductivity parameters, soil disturbance state parameters, and sensor self-test state parameters in the deployment context data, a comparison and judgment are made to obtain the trial window, excitation intensity, sampling frequency, and mechanism screening constraints, and the preset measurement mechanism is called. Within the trial window, short-term trial sampling is performed on the preset measurement mechanism according to the excitation intensity and sampling frequency to obtain trial response data. The response offset feature group is extracted in combination with the deployment context data. Based on the response offset feature group and the mechanism screening constraints, the preset measurement mechanism is subjected to availability discrimination and adaptability screening to obtain candidate measurement mechanisms.

[0010] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific steps for generating the initial acquisition strategy are as follows: By utilizing candidate measurement mechanisms and their corresponding response offset feature groups, and combining deployment context data, sorting and updating constraints and strategies are obtained to generate constraint parameters. Based on the response offset feature groups and deployment context data, competitive scores are calculated for each candidate measurement mechanism. Candidate measurement mechanisms are dynamically sorted based on competition scores and ranking update constraints to form a mechanism priority sequence. The sampling frequency, excitation intensity, sampling duration, and resampling trigger conditions are obtained through the mechanism priority sequence and strategy to generate constraint parameters, and an initial acquisition strategy is generated.

[0011] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific steps for obtaining the correction prior and forming a virtual correction strategy package are as follows: The virtual state input items are obtained by aligning and normalizing the initial acquisition strategy, response offset feature group and deployment context data; A virtual operating condition state is established based on the virtual state input items, and virtual operating condition iteration is performed to generate prior candidate values. The virtual operating condition iteration is terminated when the virtual operating condition state and the prior candidate value meet the preset convergence condition, and the prior candidate value is associated and encapsulated as a correction prior to form a virtual correction strategy package.

[0012] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific process of constructing the response quality decision model is as follows: The virtual correction strategy package is invoked to control the smart sensor to perform formal real-time sampling at the corresponding node and depth layer to obtain the original observation sequence; The hierarchical structure is constructed through an input layer, a time-series response representation layer, a working condition constraint representation layer, a gated fusion layer, a parameter decision output layer, and a quality verification layer. The input layer receives the original observation sequence, deployment context data, response offset feature group and correction prior, and inputs them into the time series response representation layer and the operating condition constraint representation layer respectively to extract response evolution features and operating condition constraint features. The gating fusion layer matches and fuses response evolution features and operating condition constraints according to node identifiers, depth level identifiers, and time identifiers. The parameter decision output layer outputs noise suppression parameters, anomaly removal thresholds, and drift compensation parameters. The quality verification layer performs output verification and parameter calibration according to the preset effective range to form a response quality decision model.

[0013] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific process of the output correction parameters is as follows: The original observation sequence and the correction prior are matched in time windows according to node identifier, depth layer identifier and time identifier, and the deployment context data and response offset feature group are simultaneously associated to form the inference input; The inference input is processed according to the hierarchical structure of the response quality decision model. The noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters are validated and boundary constraints are corrected through the quality verification layer to obtain the correction parameters.

[0014] As a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific process of forming the calibration observation sequence is as follows: Based on the correction parameters, the corresponding original observation sequence, correction prior, deployment context data and response offset feature group are read, and time windows are aligned according to node identifier, depth layer identifier and time identifier to form a joint correction input sequence. The joint calibration input sequence is traversed in chronological order. Anomaly suppression paths are determined based on anomaly removal thresholds. For the current sampled content of the original observation sequence that has not triggered anomaly suppression paths, noise suppression and drift compensation processing are performed. For the current sampled content of the original observation sequence that has triggered anomaly suppression paths, prior constraint substitution correction processing is performed to obtain the calibrated observation values ​​at each time step. The calibration observations are restored in their original chronological order to form a calibration observation sequence.

[0015] In a preferred embodiment of the high moisture content soil moisture content detection method based on intelligent sensors described in this invention, the specific process of forming the determination data is as follows: Based on the calibration observation sequence and its corresponding region identifier, node identifier, depth layer identifier, time identifier, deployment context data and response offset feature group, the multi-node calibration observation sequence in the same region is grouped by region identifier and time identifier to form a dataset to be aligned; Perform cross-sensor time alignment, cross-depth layer alignment, and cross-time window sliding alignment on the dataset to be aligned, and calculate the confidence weights of the aligned calibrated observation sequences to form the inference input package; The inference input package is input into the preset water content inference mechanism, which outputs the water content estimate, the estimate confidence level and the environmental sensitivity index, and associates and encapsulates them with the corresponding area identifier, node identifier, depth layer identifier and time identifier to form the judgment data.

[0016] The beneficial effects of this invention are as follows: by generating calibration priors through virtual working conditions, the adaptability and calibration accuracy of the detection process in high-moisture-content soil scenarios are improved; the initial acquisition strategy, response offset feature group, and deployment context data are uniformly coupled and continuously iterated and converged under virtual working conditions, so that the obtained calibration priors can accurately reflect the response change trend and calibration requirements under the current node, current layer, and current time window; the virtual calibration strategy package formed based on the calibration priors can provide continuous and consistent prior support for formal sampling, parameter decision-making, anomaly suppression, and drift compensation, enhancing the stability, pertinence, and adaptive adjustment capability of observation sequence calibration processing under time-varying environments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for detecting the moisture content of high-moisture-content soil based on intelligent sensors.

[0019] Figure 2 This is a flowchart for screening measurement mechanisms.

[0020] Figure 3 This is a flowchart for virtual operating condition iteration.

[0021] Figure 4 A flowchart for responding to quality decisions. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for detecting the moisture content of high-moisture-content soil based on a smart sensor, comprising the following steps: S1: Based on intelligent sensors, establish spatiotemporal stratigraphic identification information in the high moisture content soil area to be tested, collect operational context parameters, and form deployment context data.

[0026] S1.1: Based on the corresponding region identifier, node identifier, spatial location identifier, depth layer identifier, and time identifier, perform integrity verification and consistency matching to obtain spatiotemporal layer identifier information.

[0027] Specifically, based on the area identifier, node identifier, spatial location identifier, depth layer identifier, and time identifier corresponding to the current smart sensor, each identifier field is checked for missing, empty, duplicate, or abnormal format. Then, cross-checking is performed according to the correspondence between the same sensor node, the same spatial location, the same depth layer, and the same time record to confirm that there are no mismatches, cross-regional issues, layer conflicts, or time misalignments among the identifiers. All the identifiers that pass the verification and match are associated and integrated as spatiotemporal layer identifier information.

[0028] It should be noted that the spatiotemporal layer identification information includes region identifier, node identifier, spatial location identifier, depth layer identifier, and time identifier.

[0029] S1.2: Perform time alignment, field normalization, and outlier screening on the runtime context parameters, and associate and encapsulate them with the corresponding spatiotemporal layer identifier information to form deployment context data.

[0030] Specifically, the runtime context parameters are time-aligned according to a unified time base, and each parameter is mapped to the same sampling time or the same sampling time window. The parameter name, data format, and value type are normalized according to preset field rules. Data that exceeds the reasonable range, has abnormal mutations, missing data, or has logical conflicts is screened for outliers. The runtime context parameters that are retained after screening are associated and encapsulated with the corresponding spatiotemporal layer identifier information according to the same node, the same layer, and the same time record to form deployment context data.

[0031] It should be noted that the operating context parameters include burial depth parameters, installation posture parameters, soil disturbance status parameters, ambient temperature parameters, electrical conductivity parameters, and sensor self-test status parameters.

[0032] The preset field rules are formed by pre-setting the field name, field order, data type, value format, null value marking method, exception marking method, and time association method according to the parameter type and associated encapsulation requirements of each running context parameter.

[0033] The reasonable range is preset based on the statistical distribution of historical operating context parameters, the allowable range of sensor specifications, or the requirements of on-site deployment.

[0034] S2: Perform short-term trial sampling based on deployment context data, screen candidate measurement mechanisms and extract response offset feature groups, perform competitive scoring and dynamic ranking of candidate measurement mechanisms, and generate an initial acquisition strategy.

[0035] S2.1: Based on the environmental temperature parameters, electrical conductivity parameters, soil disturbance state parameters and sensor self-test state parameters in the deployment context data, make comparisons and judgments, obtain the trial window, excitation intensity, sampling frequency and mechanism screening constraints, and call the preset measurement mechanism.

[0036] Specifically, the ambient temperature parameter, electrical conductivity parameter, soil disturbance state parameter, and sensor self-test state parameter corresponding to the current layer of the current node are extracted from the deployment context data. The ambient temperature parameter and soil disturbance state parameter are compared with the preset window mapping interval to obtain the trial window length, the electrical conductivity parameter is compared with the preset excitation mapping interval to obtain the excitation intensity, and the ambient temperature parameter and sensor self-test state parameter are compared with the preset sampling mapping interval to obtain the sampling frequency.

[0037] Meanwhile, the ambient temperature parameters, conductivity parameters, soil disturbance state parameters, and sensor self-test state parameters are compared with the temperature application range, conductivity interference tolerance, disturbance adaptation range, and operational health requirements corresponding to each preset measurement mechanism. When a preset measurement mechanism meets all the corresponding requirements, it is determined that the preset measurement mechanism meets the mechanism screening constraints; otherwise, it is determined that the preset measurement mechanism does not meet the mechanism screening constraints.

[0038] S2.2: Within the trial window, perform short-term trial sampling on the preset measurement mechanism according to the excitation intensity and sampling frequency to obtain trial response data. Combine the deployment context data to extract response offset feature groups. Based on the response offset feature groups and mechanism screening constraints, perform availability discrimination and adaptability screening on the preset measurement mechanism to obtain candidate measurement mechanisms.

[0039] Specifically, within the trial window, each preset measurement mechanism is driven to perform short-term trial sampling according to the obtained excitation intensity and sampling frequency, continuously acquiring the corresponding trial response data. The trial response data is combined with the deployment context data. The response initial offset is extracted by the difference between the initial observation value and the baseline value, the fluctuation change is extracted by the slope of the adjacent sampling points, the saturation trend is extracted by the trend of the response amplitude change, the recovery hysteresis is extracted by the recovery time difference after the excitation ends, and the stability is extracted by the sliding variance, forming a response offset feature group. Based on whether the response offset feature group meets the mechanism screening constraints, each preset measurement mechanism is used for availability discrimination and adaptability screening. The measurement mechanisms that have effective response and adaptability under the current high water content soil conditions are retained as candidate measurement mechanisms.

[0040] S2.3: Utilize candidate measurement mechanisms and their corresponding response offset feature groups, combine deployment context data to obtain sorting update constraints and strategy to generate constraint parameters, and calculate the competitive score value for each candidate measurement mechanism based on the response offset feature groups and deployment context data.

[0041] Specifically, each candidate measurement mechanism and its corresponding response offset feature group are read. Combined with the environmental temperature, conductivity, soil disturbance state and sensor self-test state parameters reflecting the current working conditions in the deployment context data, the sorting update constraints and strategy generation constraint parameters that limit the mechanism sorting adjustment range and the acquisition strategy output boundary are extracted. Taking the response offset feature group of each candidate measurement mechanism as the core evaluation object, the response stability, saturation risk, drift risk and environmental adaptability of each candidate measurement mechanism under high water content soil conditions are comprehensively calculated in combination with the current deployment context to obtain the competitive score value corresponding to each candidate measurement mechanism.

[0042] The competition score is calculated using the following formula: ; In the formula, Indicates the first The competitive score values ​​corresponding to each candidate measurement mechanism. Indicates the first The response stability characterization values ​​of each candidate measurement mechanism Indicates the first Environmental adaptability characterization values ​​for each candidate measurement mechanism Indicates the first Saturation risk characterization value of each candidate measurement mechanism Indicates the first Drift risk characterization values ​​for each candidate measurement mechanism Indicates a saturation risk indicator. This indicates a drift risk indicator.

[0043] Among them, the response stability characterization value is obtained by comprehensive normalization calculation based on the fluctuation amplitude, number of local mutations, recovery speed and stability degree in the test response data; the environmental adaptability characterization value is obtained by normalization calculation based on the matching degree between the environmental temperature parameters, conductivity parameters, soil disturbance state parameters and sensor self-test state parameters in the deployment context data and the applicable working conditions of the candidate measurement mechanism; the saturation risk characterization value is obtained by normalization calculation based on the response amplitude approaching the limit, output plateauing, sensitivity decrease and saturation tendency degree in the test response data; and the drift risk characterization value is obtained by normalization calculation based on the response start offset, baseline change trend, offset accumulation degree and environmental disturbance sensitivity in the test response data.

[0044] In the formula for competitive score , , and All are dimensionless characterization values ​​obtained after normalization. , , and Since it is a dimensionless quantity, therefore It is a dimensionless evaluation, which meets the requirement of dimensionless measurement.

[0045] S2.4: Based on the competition score and ranking update constraints, perform dynamic ranking of candidate measurement mechanisms to form a mechanism priority sequence. Obtain the sampling frequency, excitation intensity, sampling duration and resampling trigger conditions through the mechanism priority sequence and strategy to generate constraint parameters, and generate the initial acquisition strategy.

[0046] Specifically, the candidate measurement mechanisms are ranked according to their competitive scores. The ranking and update constraints are used to restrict abnormal jumps, unstable mechanisms that may jump the queue, or mechanisms that do not meet the requirements of the current working conditions. This results in a mechanism priority sequence that reflects the priority order under the current high water content soil conditions. Based on the measurement mechanisms and strategies with higher priority in the mechanism priority sequence, constraint parameters are generated to obtain the matching sampling frequency, excitation intensity, sampling duration, and resampling trigger conditions. This generates the initial acquisition strategy for the current node and current layer.

[0047] Among them, abnormal jumps are determined by comparing whether the change in the competitive score value of the same candidate measurement mechanism in adjacent sorting rounds exceeds a preset change threshold. Low-stability mechanism insertion is determined by judging whether the stability of the response offset feature group corresponding to the candidate measurement mechanism is lower than a preset stability threshold and the sorting position suddenly increases. The mechanism front-end situation that does not meet the current working condition requirements is determined by checking whether the candidate measurement mechanism violates the mechanism screening constraints but is still located at the beginning of the mechanism priority sequence.

[0048] The preset change threshold is preset based on the statistical distribution of the change range of the competition score value in adjacent ranking rounds in the historical trial response data, and is usually taken as 1.5 to 3 times the average of the historical change range; the preset stability threshold is preset based on the normalized value range of the stability index in the response offset feature group, and is usually selected between the normalized range of 0.2 and 0.5.

[0049] S3: Construct a virtual operating condition state based on the initial acquisition strategy, response offset feature group and deployment context data, execute virtual operating condition iteration, obtain correction priors and form a virtual correction strategy package.

[0050] S3.1: The virtual state input item is obtained by aligning and normalizing the initial acquisition strategy, response offset feature group and deployment context data.

[0051] Specifically, based on the initial acquisition strategy, response offset feature group, and read deployment context data, each item of the initial acquisition strategy, response offset feature group, and deployment context data is checked according to node identifier, depth layer identifier, and time identifier. Records belonging to the same node, the same depth layer, and the same sampling time or the same sampling time window are screened out. Normalization processing is performed on sampling frequency, excitation intensity, sampling duration, resampling trigger condition, response start offset, fluctuation change, saturation tendency, recovery hysteresis, stability, ambient temperature parameter, conductivity parameter, soil disturbance state parameter, and sensor self-test state parameter respectively to identify the original value range of each item. Then, according to a unified scale, the contents with different dimensions and different value ranges are converted into directly comparable standardized expressions. The initial acquisition strategy, response offset feature group, and deployment context data after alignment and normalization processing are integrated in a fixed order to obtain virtual state input items.

[0052] S3.2: Establish virtual operating condition states based on virtual state input items and perform virtual operating condition iteration to generate prior candidate values.

[0053] Specifically, based on the virtual state input items, the initial acquisition strategy, response offset feature group, and the contents recorded at the same node, same depth layer, and same time in the deployment context data are written into the virtual working condition state. Virtual working condition iteration is performed by comparing the changes between the current round of virtual state input items and the previous round of virtual working condition state. In each round of virtual working condition iteration, the virtual working condition state is updated item by item in combination with environmental temperature parameters, conductivity parameters, soil disturbance state parameters, and sensor self-test state parameters. At the same time, the contents representing the current correction trend are extracted from the updated virtual working condition state as prior candidate values.

[0054] Among them, the virtual working condition state refers to the iterative state content that represents the current sampling strategy state, response offset state, and field working condition state after integrating the content corresponding to the same node, same depth layer, and same time record in the initial acquisition strategy, response offset feature group, and deployment context data.

[0055] The virtual operating condition is defined by the following formula: ; In the formula, Indicates the first Virtual operating condition status, Indicates the first The normalized sampling frequency in the virtual operating condition state. Indicates the first Normalized excitation intensity in the virtual working state of the cycle Indicates the first Normalized sampling duration in the virtual operating condition state. Indicates the first The normalized response start offset in the virtual operating condition state. Indicates the first Normalized fluctuation changes in the virtual operating state of the cycle. Indicates the first The saturation tendency after normalization in the virtual working condition. Indicates the first Recovery hysteresis after normalization in the virtual operating state. Indicates the first Normalized ambient temperature parameters in the virtual operating condition. Indicates the first Normalized conductivity parameters in the virtual operating condition. Indicates the first Normalized soil disturbance state parameters in the virtual working condition of the wheel. Indicates the first Normalized sensor self-test status parameters in the virtual operating condition. This indicates transpose.

[0056] A better approach is to construct a virtual working condition and perform virtual working condition iterations. This involves fusing and extrapolating the initial acquisition strategy, response offset feature groups, and deployment context data. During the iteration process, the virtual working condition is continuously updated, and prior candidate values ​​representing the current correction trend are extracted. This allows for the formation of a more stable and suitable correction basis in advance under complex high-moisture-content soil conditions.

[0057] S3.3: When the virtual operating condition state and the prior candidate value meet the preset convergence condition, the virtual operating condition iteration is terminated, and the prior candidate value is associated and encapsulated as a correction prior to form a virtual correction strategy package.

[0058] Specifically, based on prior candidate values, the changes in the current round of virtual working condition status and the previous round of virtual working condition status are continuously compared with the changes in the current round of prior candidate values ​​and the previous round of prior candidate values. When the virtual working condition status and prior candidate values ​​simultaneously meet the preset convergence conditions, the virtual working condition iteration is immediately terminated. The prior candidate values ​​that meet the preset convergence conditions are determined as correction priors. The correction priors are then checked item by item against the corresponding node identifier, depth layer identifier, time identifier, and initial acquisition strategy and written into a unified encapsulation content to form a virtual correction strategy package.

[0059] The preset convergence condition is obtained by setting a stable determination rule for consecutive iterations before the virtual working condition iteration begins. That is, it is pre-defined that when the virtual working condition state changes in the same direction as the prior candidate value in 3 to 5 consecutive iterations of the virtual working condition and the change amplitude continues to decrease and remains stable, the preset convergence condition is satisfied.

[0060] S4: Perform formal real-time sampling according to the virtual correction strategy package to obtain the original observation sequence. Input the original observation sequence and the correction prior into the response quality decision model, output the correction parameters, and perform correction processing on the original observation sequence based on the correction parameters to form the calibration observation sequence.

[0061] S4.1: Call the virtual correction strategy package to control the smart sensor to perform formal real-time sampling at the corresponding node and depth layer to obtain the original observation sequence.

[0062] Specifically, the virtual correction strategy package is invoked, and the node identifier, depth layer identifier, time identifier, initial acquisition strategy, and correction prior corresponding to the current sampling record are read from the virtual correction strategy package. Based on the node identifier and depth layer identifier, the smart sensor and corresponding sampling position that need to perform formal real-time sampling are located. According to the sampling frequency, excitation intensity, sampling duration, and resampling trigger conditions given in the initial acquisition strategy, the smart sensor is driven to continuously acquire observation signals at the corresponding node and depth layer. The continuous observation content generated during the acquisition process is recorded and organized in chronological order to obtain the original observation sequence that corresponds one-to-one with the node identifier, depth layer identifier, and time identifier.

[0063] It should be noted that the original observation sequence refers to the uncorrected observation content that is continuously recorded in chronological order and corresponds one-to-one with the node identifier, depth layer identifier, and time identifier when the smart sensor performs formal real-time sampling at the corresponding node and depth layer according to the initial acquisition strategy.

[0064] S4.2: A hierarchical structure is constructed through an input layer, a timing response representation layer, a working condition constraint representation layer, a gating fusion layer, a parameter decision output layer, and a quality verification layer.

[0065] Specifically, the parameterized hierarchical decision structure of the response quality decision model is composed of an input layer, a time-series response representation layer, a condition constraint representation layer, a gating fusion layer, a parameter decision output layer, and a quality verification layer. The input layer receives the original observation sequence, deployment context data, response offset feature set, and correction prior and arranges them in a fixed order. The time-series response representation layer is used to parameterize the continuously changing content in the original observation sequence. The condition constraint representation layer is used to parameterize the condition constraint content in the deployment context data, response offset feature set, and correction prior. The gating fusion layer is used to gating and fusing the continuously changing content and the condition constraint content. The parameter decision output layer is used to output noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters. The quality verification layer is used to perform range checks and boundary corrections on the noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters.

[0066] S4.3: The input layer receives the original observation sequence, deployment context data, response offset feature group and correction prior, and inputs them into the time series response characterization layer and the operating condition constraint characterization layer respectively to extract response evolution features and operating condition constraint features.

[0067] Specifically, the input layer receives the original observation sequence, deployment context data, response offset feature group, and correction prior. It then checks and organizes the original observation sequence, deployment context data, response offset feature group, and correction prior according to node identifier, depth layer identifier, and time identifier. The original observation sequence is then input into the time-series response characterization layer, where amplitude changes, fluctuations, abrupt change locations, and continuous change processes in the original observation sequence are parameterized and extracted to form response evolution features.

[0068] Simultaneously, deployment context data, response offset feature groups, and correction priors are input into the operating condition constraint characterization layer. The environmental temperature parameters, conductivity parameters, soil disturbance state parameters, and sensor self-test state parameters in the deployment context data, as well as the saturation tendency, recovery hysteresis, and stability degree in the response offset feature groups, along with the content representing the current correction trend in the correction priors, are parameterized, extracted, and correspondingly organized to form operating condition constraint features.

[0069] S4.4: The response evolution characteristics and operating condition constraint characteristics are constrained and fused through the gated fusion layer. The parameter decision output layer outputs noise suppression parameters, anomaly rejection thresholds and drift compensation parameters. The output verification and parameter calibration are performed through the quality verification layer to form a response quality decision model.

[0070] Specifically, the gated fusion layer performs positional and temporal correspondence checks on the response evolution features and operating condition constraint features according to node identifiers, depth level identifiers, and time identifiers. It retains feature combinations consistent with the current operating condition and suppresses feature combinations that conflict with the current operating condition. The parameter decision output layer performs parameter mapping on the fused content and outputs noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters. Then, the quality verification layer performs range checks and boundary corrections on the noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters in sequence to determine whether the noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters fall within the preset valid range. Output content that exceeds the preset valid range or does not match the current operating condition is truncated and pulled back, thereby obtaining the response quality decision model.

[0071] It should be noted that the preset effective range is set according to the statistical distribution of the values ​​of noise suppression parameters, anomaly rejection thresholds and drift compensation parameters in the historical calibration samples when the calibration effect is stable. The historical calibration samples are derived from the measured observation data and corresponding calibration test data under historical high water content soil scenarios. The preset effective range is usually located within the normalized range, for example, the example value is 0.05 to 0.95.

[0072] The response quality decision model is obtained through offline training or offline calibration. During offline training or offline calibration, the historical original observation sequence, deployment context data, response offset feature group and correction prior are used as input samples, and the corresponding noise suppression parameters, anomaly removal thresholds and drift compensation parameters in the historical correction process are used as output samples. The parameter mapping relationship in the time series response representation layer, operating condition constraint representation layer, gating fusion layer and parameter decision output layer is adjusted repeatedly. After the deviation between the output sample and the actual correction parameters meets the preset requirements, the parameters are fixed to form the response quality decision model.

[0073] S4.5: Match the original observation sequence with the correction prior according to node identifier, depth layer identifier and time identifier, and synchronously associate the deployment context data and response offset feature group to form the inference input.

[0074] Specifically, the node identifier, depth layer identifier, and time identifier are read from the original observation sequence, and the node identifier, depth layer identifier, and time identifier are read from the calibration prior. The original observation sequence and the calibration prior are checked for correspondence item by item. For content with consistent node identifiers, consistent depth layer identifiers, and time identifiers corresponding to the same sampling time or the same sampling time window, time window matching is performed. The deployment context data and response offset feature group corresponding to the time window matching content are read. The node identifier, depth layer identifier, and time identifier in the deployment context data and response offset feature group are checked for consistency with the time window matching content. The original observation sequence, calibration prior, deployment context data, and response offset feature group that pass the check are integrated into the same record in a fixed order to form the inference input.

[0075] S4.6: The inference input is processed according to the hierarchical structure of the response quality decision model. The noise suppression parameters, anomaly rejection thresholds and drift compensation parameters are validated and boundary constraints are corrected through the quality verification layer to obtain the correction parameters.

[0076] Specifically, the inference input is sequentially fed into the input layer, temporal response representation layer, operating condition constraint representation layer, gating fusion layer, parameter decision output layer, and quality verification layer of the response quality decision model. The temporal response representation layer processes the continuously changing content in the original observation sequence. The operating condition constraint representation layer processes the operating condition constraint content in the correction prior, deployment context data, and response offset feature group. The gating fusion layer fuses the continuously changing content and operating condition constraint content according to node identifier, depth layer identifier, and time identifier. The parameter decision output layer outputs noise suppression parameters, anomaly removal thresholds, and drift compensation parameters. The quality verification layer performs validity verification and boundary constraint correction on the noise suppression parameters, anomaly removal thresholds, and drift compensation parameters to obtain the correction parameters.

[0077] It should be noted that continuous changes refer to the amplitude fluctuations, wave continuation, slope of change, abrupt transitions, and continuous change relationships between adjacent sampling points presented in chronological order in the original observation sequence; operating condition constraints refer to the constraint information in the correction prior, deployment context data, and response offset feature group that reflects environmental temperature parameters, conductivity parameters, soil disturbance state parameters, sensor self-test state parameters, saturation tendency, recovery hysteresis, and stability.

[0078] Noise suppression parameters are correction parameters that limit the suppression intensity of high-frequency fluctuations, local spikes, and short-term jitter in the original observation sequence. Anomaly removal thresholds are discrimination parameters that determine whether there are sudden deviations or discontinuities in the current sampling content of the original observation sequence and whether it enters the anomaly suppression path. Drift compensation parameters are correction parameters that correct the baseline offset, slow drift, and trend offset amplitude that accumulate over time in the original observation sequence.

[0079] S4.7: Based on the correction parameters, read the corresponding original observation sequence, correction prior, deployment context data and response offset feature group, and perform time window alignment according to node identifier, depth layer identifier and time identifier to form a joint correction input sequence.

[0080] Specifically, the noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters corresponding to the current sampling record are read, and then the original observation sequence, correction prior, deployment context data, and response offset feature group of the same node, the same depth layer, and the same sampling time or the same sampling time window are read synchronously.

[0081] The calibration parameters, original observation sequence, calibration prior, deployment context data, and response offset feature group are checked item by item according to node identifier, depth layer identifier, and time identifier. The calibration parameters, original observation sequence, calibration prior, deployment context data, and response offset feature group are checked to see if they come from the same smart sensor node, if they correspond to the same depth layer, and if they fall into the same sampling time or the same sampling time window.

[0082] The calibration parameters, original observation sequence, calibration prior, deployment context data, and response offset feature group that meet the requirements of consistent node identifier, consistent depth layer identifier, and consistent time identifier are written into the same record in a fixed order to form a joint calibration input sequence.

[0083] S4.8: Traverse the joint calibration input sequence in chronological order, determine the anomaly suppression path based on the anomaly removal threshold, perform noise suppression processing and drift compensation processing on the current sampled content of the original observation sequence that has not triggered the anomaly suppression path, and perform prior constraint substitution correction processing on the current sampled content of the original observation sequence that has triggered the anomaly suppression path to obtain the calibration observation value at each time.

[0084] Specifically, the records in the joint correction input sequence are read one by one in chronological order. The current sampled content of the original observation sequence in each record is compared with the anomaly rejection threshold. It is checked whether there are abrupt deviations or discontinuities in the current sampled content of the original observation sequence that exceed the anomaly rejection threshold. The anomaly suppression path is determined based on the checked content. For records that have not triggered the anomaly suppression path, the high-frequency fluctuations and local spikes in the current sampled content of the original observation sequence are suppressed according to the noise suppression parameters. Then, the offset formed by the accumulation of time in the current sampled content of the original observation sequence is corrected according to the drift compensation parameters.

[0085] For records that trigger anomaly suppression paths, the current sampled content of the original observation sequence is not directly retained. Instead, the deviation between the current sampled content of the original observation sequence and the correction prior is compared. The abnormal deviation part is weighted and replaced by the content in the correction prior that represents the current correction trend according to the deviation magnitude between the current sampled content and the correction prior. The content after noise suppression processing, drift compensation processing or prior constraint replacement correction processing is used as the calibration observation value at the corresponding time.

[0086] Among them, mutation bias represents a sudden increase or decrease in the threshold of a single sampling time relative to the previous sampling time, and discontinuity bias represents an anomaly in the threshold where the current sampling time loses its continuous connection with both the previous and next adjacent sampling times.

[0087] When the change in the current sample content of the original observation sequence relative to the previous sample content does not exceed the anomaly removal threshold, and the current sample content of the original observation sequence maintains a continuous change relationship with the adjacent sample content, it is determined that the anomaly suppression path has not been triggered. When the change in the current sample content of the original observation sequence relative to the previous sample content exceeds the anomaly removal threshold, or when there is a sudden deviation or discontinuity deviation between the current sample content of the original observation sequence and the adjacent sample content that exceeds the anomaly removal threshold, it is determined that the anomaly suppression path has been triggered.

[0088] For the first sampling time, only the change magnitude of the current sampled content and the next sampling time is compared to determine whether the anomaly rejection threshold is exceeded to determine whether the anomaly suppression path is triggered; for the last sampling time, only the change magnitude of the current sampled content and the previous sampling time is compared to determine whether the anomaly rejection threshold is exceeded to determine whether the anomaly suppression path is triggered.

[0089] A better approach is to use a fixed filter, fixed anomaly threshold, or uniform drift compensation rule for single-path correction processing of the original observation sequence. Instead, anomaly suppression paths are determined based on anomaly removal thresholds. Then, noise suppression and drift compensation processing are performed on records that do not trigger anomaly suppression paths, and prior constraint substitution correction processing is performed on records that trigger anomaly suppression paths. This improves the pertinence and adaptability of correction processing in high-moisture-content soil scenarios.

[0090] S4.9: Restore the calibration observations in their original time sequence to form a calibration observation sequence.

[0091] Specifically, the calibration observations obtained at each time point are rearranged according to their corresponding time markers to restore the time sequence consistent with the original sampling order. Calibration observations that have the same sampling time or the same sampling time window are merged and continuously connected to form a calibration observation sequence.

[0092] S5: Based on the calibration observation sequence, perform data alignment and confidence weighting across sensors, depth layers and time windows, and output water content estimates, estimated confidence levels and environmental sensitivity indicators through a preset water content inference mechanism to form judgment data.

[0093] S5.1: Based on the calibration observation sequence and its corresponding region identifier, node identifier, depth layer identifier, time identifier, deployment context data and response offset feature group, group the multi-node calibration observation sequences in the same region according to the region identifier and time identifier to form a dataset to be aligned.

[0094] Specifically, the calibration observation sequence corresponding to the current node is read, and the region identifier, node identifier, depth layer identifier, time identifier, deployment context data, and response offset feature group corresponding to each calibration observation sequence are extracted simultaneously. The calibration observation sequence is checked item by item according to the region identifier and time identifier. Multi-node calibration observation sequences with the same region identifier and time identifier belonging to the same sampling time time or the same sampling time window are merged into the same group. At the same time, the corresponding node identifier, depth layer identifier, deployment context data, and response offset feature group within each group are retained to form a dataset to be aligned.

[0095] The dataset to be aligned refers to the set of multi-node calibration observation sequences with the same regional identifier and time identifier belonging to the same sampling time or the same sampling time window, together with the corresponding node identifier, depth layer identifier, deployment context data and response offset feature group.

[0096] S5.2: Perform cross-sensor time alignment, cross-depth layer alignment, and cross-time window sliding alignment on the dataset to be aligned, and calculate the confidence weight of the aligned calibrated observation sequence to form the inference input package.

[0097] Specifically, the multi-node calibration observation sequences in the dataset to be aligned are first aligned across sensors according to time identifiers, so that the calibration observation contents of different nodes in the same area at the same sampling time or the same sampling time window are mapped to the same time position. Then, they are aligned across depth layers according to depth layer identifiers, so that the correspondence between different depth layers is established at the same time position.

[0098] The calibration observations in adjacent sampling time windows are continuously aligned across time windows using a sliding alignment method. The consistency of time, the consistency of layer correspondence, the continuity of adjacent time windows, and the degree of matching with deployment context data and response offset feature groups among the calibration observation sequences are checked one by one. The confidence weight of each calibration observation sequence is calculated based on the check results. The aligned calibration observation sequences and their corresponding confidence weights are then integrated in a fixed order to form the inference input package.

[0099] S5.3: Input the inference input package into the preset water content inference mechanism, output the water content estimate, the estimate confidence level and the environmental sensitivity index, and associate and encapsulate them with the corresponding area identifier, node identifier, depth layer identifier and time identifier to form the judgment data.

[0100] Specifically, the calibration observation sequence and corresponding confidence weights in the inference input package are sent to the preset water content inference mechanism in a fixed order. The preset water content inference mechanism calculates the observation content representing the current water content in the calibration observation sequence item by item, and outputs the water content estimate, estimate confidence and environmental sensitivity index of the current region, current node and current depth layer at the current sampling time or the current sampling time window in combination with the corresponding confidence weight. The water content estimate, estimate confidence and environmental sensitivity index are checked with the corresponding region identifier, node identifier, depth layer identifier and time identifier item by item, and then associated and encapsulated as the same record to form the judgment data.

[0101] The preset moisture content inference mechanism uses the moisture content fitting function obtained from offline calibration. It first performs weighted calculations on the calibration observation sequence in the inference input package according to the confidence weight, and obtains the weighted observation value at the current sampling time or the current sampling time window.

[0102] In summary, this invention improves the adaptability and accuracy of the detection process in high-moisture-content soil scenarios by iteratively generating calibration priors under virtual working conditions; it unifies and couples the initial acquisition strategy, response offset feature group, and deployment context data, and continuously iterates and converges under virtual working conditions, so that the obtained calibration priors can accurately reflect the response change trends and calibration requirements under the current node, current layer, and current time window; the virtual calibration strategy package formed based on the calibration priors can provide continuous and consistent prior support for formal sampling, parameter decision-making, anomaly suppression, and drift compensation, enhancing the stability, specificity, and adaptive adjustment capability of observation sequence calibration processing under time-varying environments.

[0103] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting the moisture content of high-moisture-content soil based on intelligent sensors, characterized in that, include: Based on intelligent sensors, spatiotemporal stratigraphic identification information is established in the high moisture content soil area to be tested, and operational context parameters are collected to form deployment context data. Short-term trial sampling is performed based on deployment context data to screen candidate measurement mechanisms and extract response offset feature groups. The candidate measurement mechanisms are then competitively scored and dynamically ranked to generate an initial acquisition strategy. Based on the initial acquisition strategy, response offset feature group and deployment context data, a virtual operating condition state is constructed, virtual operating condition iteration is performed, correction prior is obtained and a virtual correction strategy package is formed. Formal real-time sampling is performed according to the virtual correction strategy package to obtain the original observation sequence. The original observation sequence is then input into the correction prior response quality decision model to output correction parameters. Based on the correction parameters, the original observation sequence is corrected to form a calibrated observation sequence. Based on the calibration observation sequence, data alignment and confidence weighting are performed across sensors, depth layers, and time windows. The water content estimate, estimated confidence level, and environmental sensitivity index are output through a preset water content inference mechanism to form judgment data.

2. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 1, characterized in that, The spatiotemporal layer identification information includes region identifier, node identifier, spatial location identifier, depth layer identifier, and time identifier.

3. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 1, characterized in that, The specific steps for forming deployment context data are as follows: Based on the corresponding region identifier, node identifier, spatial location identifier, depth layer identifier, and time identifier, perform integrity verification and consistency matching to obtain spatiotemporal layer identifier information; The runtime context parameters are aligned to the execution time, normalized to the fields, and screened for outliers. They are then associated and encapsulated with the corresponding spatiotemporal layer identifiers to form deployment context data.

4. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 1, characterized in that, The specific steps for screening candidate measurement mechanisms and extracting response offset feature groups are as follows: Based on the environmental temperature parameters, electrical conductivity parameters, soil disturbance state parameters, and sensor self-test state parameters in the deployment context data, a comparison and judgment are made to obtain the trial window, excitation intensity, sampling frequency, and mechanism screening constraints, and the preset measurement mechanism is called. Within the trial window, short-term trial sampling is performed on the preset measurement mechanism according to the excitation intensity and sampling frequency to obtain trial response data. The response offset feature group is extracted in combination with the deployment context data. Based on the response offset feature group and the mechanism screening constraints, the preset measurement mechanism is subjected to availability discrimination and adaptability screening to obtain candidate measurement mechanisms.

5. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 4, characterized in that, The specific steps for generating the initial acquisition strategy are as follows: By utilizing candidate measurement mechanisms and their corresponding response offset feature groups, and combining deployment context data, sorting and updating constraints and strategies are obtained to generate constraint parameters. Based on the response offset feature groups and deployment context data, competitive scores are calculated for each candidate measurement mechanism. Candidate measurement mechanisms are dynamically sorted based on competition scores and ranking update constraints to form a mechanism priority sequence. The sampling frequency, excitation intensity, sampling duration, and resampling trigger conditions are obtained through the mechanism priority sequence and strategy to generate constraint parameters, and an initial acquisition strategy is generated.

6. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 1, characterized in that, The specific steps for obtaining the correction prior and forming the virtual correction strategy package are as follows: The virtual state input items are obtained by aligning and normalizing the initial acquisition strategy, response offset feature group and deployment context data; A virtual operating condition state is established based on the virtual state input items, and virtual operating condition iteration is performed to generate prior candidate values. The virtual operating condition iteration is terminated when the virtual operating condition state and the prior candidate value meet the preset convergence condition, and the prior candidate value is associated and encapsulated as a correction prior to form a virtual correction strategy package.

7. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 1, characterized in that, The specific process of constructing the response quality decision model is as follows: The virtual correction strategy package is invoked to control the smart sensor to perform formal real-time sampling at the corresponding node and depth layer to obtain the original observation sequence; The hierarchical structure is constructed through an input layer, a time-series response representation layer, a working condition constraint representation layer, a gated fusion layer, a parameter decision output layer, and a quality verification layer. The input layer receives the original observation sequence, deployment context data, response offset feature group and correction prior, and inputs them into the time series response representation layer and the operating condition constraint representation layer respectively to extract response evolution features and operating condition constraint features. The gating fusion layer matches and fuses response evolution features and operating condition constraints according to node identifiers, depth level identifiers, and time identifiers. The parameter decision output layer outputs noise suppression parameters, anomaly removal thresholds, and drift compensation parameters. The quality verification layer performs output verification and parameter calibration according to the preset effective range to form a response quality decision model.

8. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 7, characterized in that, The specific process for determining the output correction parameters is as follows: The original observation sequence and the correction prior are matched in time windows according to node identifier, depth layer identifier and time identifier, and the deployment context data and response offset feature group are simultaneously associated to form the inference input; The inference input is processed according to the hierarchical structure of the response quality decision model. The noise suppression parameters, anomaly rejection thresholds, and drift compensation parameters are validated and boundary constraints are corrected through the quality verification layer to obtain the correction parameters.

9. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 8, characterized in that, The specific process for forming the calibration observation sequence is as follows: Based on the correction parameters, the corresponding original observation sequence, correction prior, deployment context data and response offset feature group are read, and time windows are aligned according to node identifier, depth layer identifier and time identifier to form a joint correction input sequence. The joint calibration input sequence is traversed in chronological order. Anomaly suppression paths are determined based on anomaly removal thresholds. For the current sampled content of the original observation sequence that has not triggered anomaly suppression paths, noise suppression and drift compensation processing are performed. For the current sampled content of the original observation sequence that has triggered anomaly suppression paths, prior constraint substitution correction processing is performed to obtain the calibrated observation values ​​at each time step. The calibration observations are restored in their original chronological order to form a calibration observation sequence.

10. The method for detecting the moisture content of high-moisture-content soil based on intelligent sensors as described in claim 1, characterized in that, The specific process for generating the determination data is as follows: Based on the calibration observation sequence and its corresponding region identifier, node identifier, depth layer identifier, time identifier, deployment context data and response offset feature group, the multi-node calibration observation sequence in the same region is grouped by region identifier and time identifier to form a dataset to be aligned; Perform cross-sensor time alignment, cross-depth layer alignment, and cross-time window sliding alignment on the dataset to be aligned, and calculate the confidence weights of the aligned calibrated observation sequences to form the inference input package; The inference input package is input into the preset water content inference mechanism, which outputs the water content estimate, the estimate confidence level and the environmental sensitivity index, and associates and encapsulates them with the corresponding area identifier, node identifier, depth layer identifier and time identifier to form the judgment data.

Images