Irrigation area water regime monitoring method, device and system using sensor technology

Abstract

The irrigation water regime monitoring method, device and system using sensor technology belong to the technical field of irrigation water regime regulation systems, and comprise the following steps: S1, receiving model parameter updates uploaded by each irrigation sub-region under a federal learning framework, and iteratively updating a central side irrigation digital twin model with the model parameter updates; S2, obtaining sensor measured values and artificial intervention data derived from system logs; S3, calculating a global average distortion index and an artificial intervention rate; the global average distortion index is used to represent the deviation between the digital twin model and the physical world reflected by the sensor measured values, and the artificial intervention rate is used to represent the degree of decision conflict based on artificial intervention data statistics; S4, based on a preset weight, combining the global average distortion index and the artificial intervention rate, the man-machine system trust entropy is calculated out, and the present application provides a benchmark for sensor measured values which is independent of the sensor state itself and has a clear physical meaning.

Description

Technical Field

[0001] The present invention relates to the technical field of irrigation area water regime regulation systems, and in particular to an irrigation area water regime monitoring method, equipment and system using sensor technology. Background Art

[0002] Existing irrigation district water regime monitoring generally relies on sensor networks. By deploying sensors for water level, flow, and soil moisture, combined with wireless communication technology and back-end data platforms, remote monitoring and management of water resources are achieved. This technological system aims to improve the level of precision in irrigation management. Existing solutions typically focus on building high-fidelity digital twin models and pursuing globally optimal scheduling strategies. However, when faced with complex real-world scenarios, existing technologies have significant shortcomings. They rely heavily on centralized, high-quality sensor data, but due to cross-domain management constraints, raw data aggregation is difficult. These systems do not account for non-ideal conditions such as aging physical facilities and systematic sensor failures. In particular, when sensor data is systematically distorted due to unmodeled physical phenomena, such as soil salinization, decisions based on erroneous data can conflict sharply with the practical experience of frontline managers. This conflict can gradually erode trust in machines, ultimately leading to the abandonment of intelligent systems and a return to traditional manual methods. Existing technologies lack a mechanism to quantify the state of trust between humans and machines and proactively adjust control strategies when trust deteriorates.

[0003] The technical problem to be solved by the present invention is to provide an irrigation scheduling method, equipment and system that can dynamically evaluate and calibrate the system trust and adaptively switch control strategies when facing systematic data distortion and human-machine decision conflicts, so as to avoid trust collapse and ensure the long-term stable operation of the system under deep uncertainty.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, device and system for monitoring water conditions in an irrigation area using sensor technology to solve the problems raised in the above background technology.

[0006] The technical solution of the present invention comprises the following steps:

[0007] S1. Receive model parameter updates uploaded by each irrigation sub-region under the federated learning framework, and iteratively update the digital twin model of the center-side irrigation area using the model parameter updates.

[0008] S2. Obtain sensor measured values ​​and manual intervention data from system logs;

[0009] S3. Calculate the global average distortion index and human intervention rate. The global average distortion index is used to characterize the deviation between the digital twin model and the physical world reflected by the sensor measurements. The human intervention rate is used to characterize the degree of decision conflict based on human intervention data statistics.

[0010] S4. Based on the preset weights, combined with the global average distortion index and the manual intervention rate, the trust entropy of the human-machine system is calculated;

[0011] S5. Compare the trust entropy of the human-machine system with the preset trust entropy safety threshold and trust entropy danger threshold, and adaptively switch the irrigation scheduling mode based on the comparison result.

[0012] Preferably, the step of calculating the global average distortion index in S3 specifically includes:

[0013] Based on the water balance principle and meteorological monitoring data, the theoretical change in soil volume water content of the plot within the time window is calculated;

[0014] A normalization factor that only considers the contribution of total water inflow is calculated. The theoretical change is combined with the actual change in water content measured by the sensor, and the normalization factor is used to calculate the sensor data distortion index for each plot. The sensor data distortion index of all plots is spatially smoothed and averaged to generate a global average distortion index.

[0015] Preferably, the step of calculating the manual intervention rate in S3 specifically includes:

[0016] Obtain the total number of decision-making intervention operations performed by managers within a preset time window from the system log;

[0017] From the system log, obtain the total number of scheduling instructions issued by the system within the same time window;

[0018] The manual intervention rate is calculated by dividing the total number of decision intervention operations by the total number of scheduling instructions.

[0019] Preferably, S4 specifically includes:

[0020] The global average distortion index and the human intervention rate are weighted using the weight coefficients determined by the expert calibration method and the historical data regression method, and the weighted results are summed to generate the trust entropy of the human-machine system.

[0021] Preferably, S5 specifically includes:

[0022] When the trust entropy of the human-machine system is less than the trust entropy safety threshold, the system operates in the global optimal scheduling mode;

[0023] When the trust entropy of the human-machine system is greater than or equal to the trust entropy safety threshold and less than the trust entropy danger threshold, the system switches to the local enhanced decision-making mode;

[0024] When the trust entropy of the human-machine system is greater than or equal to the trust entropy danger threshold, the system switches to human-machine co-driving mode.

[0025] Preferably, in the human-machine co-driving mode, it also includes:

[0026] The system marks the associated data sources that cause the trust entropy to exceed the limit to guide operation and maintenance personnel to conduct physical inspections and repairs; when the trust entropy of the human-machine system falls below the trust entropy safety threshold due to repairs or scheduling mode adjustments, the system returns to the global optimal scheduling mode, forming a negative feedback closed loop.

[0027] Irrigation area water condition monitoring equipment using sensor technology, including:

[0028] processor;

[0029] A memory storing a computer program.

[0030] Irrigation area water condition monitoring system using sensor technology, including:

[0031] The perception and execution layer is composed of sensors and intelligent control actuators deployed in the irrigation area;

[0032] The network transmission layer is responsible for transmitting data and instructions between the perception execution layer and the platform support layer;

[0033] Platform support layer.

[0034] The present invention provides an improved irrigation area water regime monitoring method, device, and system using sensor technology. Compared with the prior art, the present invention has the following improvements and advantages:

[0035] 1. This approach constructs a global average distortion index, providing a physically meaningful benchmark for sensor measurements that is independent of the sensor's state. The underlying logic is that by comparing actual sensor readings with theoretically predicted values, it can reveal systematic deviations that cannot be explained by known physical processes.

[0036] 2. The index is freed from the influence of absolute values ​​such as irrigation volume and rainfall, enabling fair comparisons across different plots and over time. This provides a solid theoretical basis for the calculation of the global average distortion index. Each manual intervention is considered a veto of the machine's decision by a human expert, providing an objective and direct quantitative basis for measuring conflict at the decision-making level.

[0037] 3. The trust entropy calculation model ensures not only a reasonable structure, but also parameters that incorporate historical experience and expert knowledge, resulting in a high degree of real-world applicability and credibility. This solution enables adaptive switching of scheduling strategies, a fundamental advancement compared to the rigid control strategies of existing technologies. This allows the system's autonomy to dynamically match its real-time credibility, pursuing global optimization when the system is in good condition, proactively narrowing the decision-making scope and reducing risk when trust decreases, and returning core decision-making power to humans when trust is severely insufficient. This avoids catastrophic decisions driven by erroneous data and significantly improves the system's resilience and security in complex real-world environments.

[0038] 4. This solution establishes a complete negative feedback loop through fault tracing and guided repair capabilities in the human-machine co-pilot mode. When the system enters the human-machine co-pilot mode, it not only passively relinquishes control but also proactively identifies associated data sources that cause excessive trust entropy, thereby guiding operations personnel to conduct precise physical inspections and repairs. This process directly reduces the global average distortion index and human intervention rate, thereby causing the trust entropy of the human-machine system to fall back. When it recovers below the safety threshold, the system can automatically return to the globally optimal scheduling mode. This design enables the entire system to evolve from a static decision-making tool into a resilient organism that can co-evolve with the environment and humans and possess a certain degree of self-repair capabilities, ensuring its vitality and sustainable adoption in long-term operation. This is also an advanced feature that is completely lacking in existing technical solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present invention will be further explained below in conjunction with the accompanying drawings and Examples:

[0040] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION

[0041] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0042] Example 1:

[0043] See also Figure 1 The present invention provides a method for monitoring water conditions in an irrigation area using sensor technology, comprising the following steps:

[0044] S1. Receive model parameter updates uploaded by each irrigation sub-region under the federated learning framework, and iteratively update the digital twin model of the center-side irrigation area using the model parameter updates.

[0045] S2. Obtain sensor measured values ​​and manual intervention data from system logs;

[0046] S3. Calculate the global average distortion index and human intervention rate. The global average distortion index is used to characterize the deviation between the digital twin model and the physical world reflected by the sensor measurements. The human intervention rate is used to characterize the degree of decision conflict based on human intervention data statistics.

[0047] S4. Based on the preset weights, combined with the global average distortion index and the manual intervention rate, the trust entropy of the human-machine system is calculated;

[0048] S5. Compare the trust entropy of the human-machine system with the preset trust entropy safety threshold and trust entropy danger threshold, and adaptively switch the irrigation scheduling mode based on the comparison result;

[0049] This embodiment describes the complete process of an adaptive irrigation scheduling method including human-machine trust calibration. The method is implemented in an irrigation scheduling system consisting of a perception execution layer, a network transmission layer, and a platform support layer.

[0050] The initial step of the method is that in step S1, in order to optimize the global model using the data of each sub-region under the premise of complying with data management constraints, a federated learning framework is introduced; the federated learning framework refers to a distributed machine learning framework that allows each sub-region computing node to use the original sensor data locally for model training, and only uploads the privacy-processed model parameter updates to the central server to ensure that the original data does not leave the local area; the irrigation district digital twin model on the central side continuously receives these parameter updates; the irrigation district digital twin model refers to a virtual model built based on the hybrid modeling principle of physical information and data-driven, used to characterize the global hydrological dynamic state of the irrigation district. Its function is to simulate and predict the water volume changes, crop water demand and other states of the irrigation district, and provide a decision-making basis for global optimal scheduling; the model iterates through the received parameter updates, thereby continuously improving its representation accuracy of the actual state of the physical irrigation district;

[0051] For example, this hybrid model can be implemented as follows: using a physical process-based soil water balance model, such as the SWAT or HYDRUS model, as the basic framework to simulate the main irrigation, rainfall, evapotranspiration, and infiltration processes; at the same time, coupling it with a data-driven time series prediction model, such as a long short-term memory network (LSTM), to use historical sensor data to learn and compensate for model deviations caused by complex nonlinear factors such as soil compaction and salt accumulation that are not explicitly included in the physical model, thereby realizing the complementary advantages of the two modeling paradigms;

[0052] In step S2, the system acquires data; this step forms the data foundation for subsequent quantitative calculations. On the one hand, the system obtains sensor-measured values ​​reflecting the physical state from soil moisture sensors and other devices deployed in each irrigation plot. On the other hand, the system continuously records and extracts human intervention data from the operation log. Human intervention data refers to the records of irrigation district managers using the control platform to cancel, modify, or force manual control of automatically generated system scheduling instructions. This data directly reflects the human experts' distrust of the system's decisions.

[0053] Based on the acquired data, in step S3, the system quantifies two core trust-influencing factors. First, it calculates the global average distortion index. The global average distortion index is a dimensionless scalar that quantifies the overall deviation between the digital twin model and physical reality at the data level. Second, it calculates the human intervention rate. The human intervention rate is a dimensionless ratio that quantifies the intensity of the conflict between human managers and the AI ​​system at the decision-making level.

[0054] In step S4, the two quantitative indicators of different dimensions are combined into a single trust evaluation indicator. This step is based on preset weights and combines the global average distortion index and the human intervention rate to calculate the trust entropy of the human-machine system. The trust entropy of the human-machine system is a comprehensive dimensionless scalar, and its physical meaning is defined as the degree of disorder or unreliability of the entire human-machine collaborative system. The higher its value, the more unstable the system state is and the lower the trust level of the human-machine system.

[0055] The final step of the method is to apply the calculated trust entropy to the actual control loop in step S5. The system compares the real-time calculated trust entropy of the human-machine system with two preset thresholds: the trust entropy safety threshold and the trust entropy danger threshold. The trust entropy safety threshold and the trust entropy danger threshold are critical values ​​set based on historical data distribution or risk simulation, and together they define the safe zone, warning zone, and danger zone of the system's trust state. Based on the comparison results, the system automatically and smoothly switches between three preset irrigation scheduling modes: the global optimal scheduling mode, the local enhanced decision-making mode, and the human-machine co-pilot mode.

[0056] This method achieves the quantitative calculation and application of the fuzzy concept of human-machine trust in irrigation scheduling systems by constructing a complete closed loop from data acquisition to model update, conflict quantification, trust assessment, and strategy switching. It also solves the fundamental problem of trust collapse caused by the lack of feedback adjustment mechanisms in existing technologies when faced with data distortion and decision conflicts. It can dynamically adjust the system's autonomy and control strategy while ensuring data privacy and data management constraints, ensuring the long-term stability and reliability of the entire intelligent irrigation system in complex and uncertain real-world scenarios, thereby improving the level of refinement of water resource management and the practical value of intelligent systems.

[0057] Compared to the data management constraints and privacy barriers inherent in centralized data processing models in existing technologies, the federated learning framework introduced in this solution achieves continuous iterative optimization of the central irrigation district digital twin model by only interacting to update model parameters, without moving the original sensor data of each irrigation sub-region. This distributed learning paradigm not only fundamentally complies with data management constraints but also achieves an organic unity of global vision and local data privacy that was previously unattainable with previous technologies, providing a feasible path for building accurate global models.

[0058] The core advancement of this invention lies in the rigorous quantitative characterization of the deviation between the physical world and the digital model, as well as the conflict between human and machine decision-making, and their integration into a unified system status evaluation indicator.

[0059] Example 2

[0060] The steps for calculating the global average distortion index in S3 specifically include:

[0061] Based on the water balance principle and meteorological monitoring data, the theoretical change in soil volume water content of the plot within the time window is calculated;

[0062] Calculate a normalization factor that only considers the contribution of total water inflow; combine the theoretical change with the actual change in water content measured by the sensor, and use the normalization factor to calculate the sensor data distortion index for each plot; perform spatial smoothing and averaging of the sensor data distortion index for all plots to generate a global average distortion index;

[0063] In this embodiment, the global average distortion index calculation process in step S3 is described. This process aims to reveal the sensor systematic distortion caused by unmodeled factors such as soil salinization by comparing the physical model with the measured data.

[0064] In order to obtain a reference that is not affected by the state of the sensor itself, the initial step of the calculation process is that the system calculates the water balance principle for any plot. In the time window Theoretical change in soil volume water content within The calculation formula is:

[0065] ;

[0066] In this formula, each parameter is derived from clear physical measurements or preset information, among which, is the theoretical change in soil volume water content, is the irrigation volume measured by the flow meter, is the rainfall depth monitored by the local weather station, The evapotranspiration rate is calculated by combining meteorological data with the well-known models such as Penman-Monteith. is the plot area, is the effective root depth of the crop, is the time window length; this calculation ensures the physical authenticity of the theoretical prediction value;

[0067] In order to avoid the numerical instability that may occur when there is no rainfall and irrigation, and to make the distortion measurement more robust, the calculation process further introduces a normalization factor that represents the water input intensity. ; This factor only considers the theoretical contribution of total water inflow to soil moisture content, and the calculation formula is:

[0068] ;

[0069] The introduction of this normalization factor enhances the robustness of the subsequent distortion index calculation and avoids the problem of the denominator being zero when the inlet and outlet water balance is approximately zero;

[0070] Based on the above theoretical calculation results, combined with the actual measurement values ​​of the sensor, the single plot can be calculated Sensor data distortion index The calculation formula is:

[0071] ;

[0072] Here By plot Soil moisture sensors in the same time window The actual change in water content measured internally is derived from the actual value measured by the sensor; is a dimensionless regularization constant set to prevent the denominator from being zero when there is no water input; : The index of the plot;

[0073] To ensure that the value of this constant has a clear engineering basis rather than an arbitrary setting, a preferred setting method is to set The value of is related to the measurement accuracy or minimum resolution of the sensor itself. For example, it can be set to the theoretical water content change value corresponding to the minimum valid reading change of a single measurement of the sensor. This not only avoids singularities in the calculation, but also ensures that only when the deviation between the prediction and the measurement, i.e. the numerator , significantly exceeds the normal measurement noise range of the sensor, the distortion index Only when the value is significantly greater than zero can the index's anti-interference ability and the clarity of its physical meaning be enhanced;

[0074] As a dimensionless relative error, if its value is persistently high, it indicates that the sensor data of the plot has a systematic deviation from the physical reality that requires attention;

[0075] The final step in the calculation process is to calculate the sensor data distortion index for all plots in the irrigation area. Perform spatial smoothing and averaging, such as arithmetic averaging or weighted averaging, to generate a global average distortion index ; It represents the overall distortion of the data level of the entire irrigation area; the calculation formula is:

[0076] ;

[0077] in, is the total number of plots within the irrigation area;

[0078] Through the above-mentioned specific implementation methods, the present invention provides a clear and reproducible calculation path for the core concept of the global average distortion index. It cleverly utilizes the physical model as an anchor point, making the detection of systematic sensor distortion no longer dependent on cross-validation of other sensors, but based on first principles of physics, making it more reliable. The introduction of the normalization factor significantly improves the robustness of the algorithm. The resulting index can effectively quantify the deviation between the digital twin model and the physical world, providing a high-quality and highly reliable data foundation for the subsequent precise calculation of trust entropy, which is a key prerequisite for achieving accurate trust assessment.

[0079] In order to overcome the problem that existing technologies cannot effectively deal with unmodeled physical phenomena, such as the systematic distortion of sensors caused by soil salinization, this invention constructs a global average distortion index. The physical meaning of this index comes from the application of the basic conservation law of water balance principle; the theoretical change of soil volume water content ,in, : irrigation volume, : rainfall depth, :area, : evapotranspiration rate, :time, The calculation of root depth is not a pure data fitting, but a deduction based on the first principles of physics. It provides a benchmark with clear physical meaning for the actual value of the sensor, which is independent of the state of the sensor itself. Its internal logic is to compare the actual reading change of the sensor. This physical theory predicts , can reveal systematic deviations that cannot be explained by known physical processes;

[0080] In order to make the measurement of this deviation robust and comparable, the scheme further introduces a normalization factor that only considers the contribution of the total water inflow ,in, : irrigation volume, : rainfall depth, :area, : Root depth; The role of this factor is to convert the absolute error into a relative error, so that the final distortion index of the plot sensor data ,in, : Measured water content changes, : Predict water content changes, : Input moisture contribution, : The regularization constant becomes a dimensionless quantity, and its physical meaning is the proportion of unexpected changes in water content contributed by unmodeled factors in the total input water. This design makes the index free from the influence of absolute magnitudes such as irrigation volume and rainfall, and enables fair comparisons across different plots and at different times, thus providing a global average distortion index. The calculation provides a solid theoretical basis.

[0081] Example 3

[0082] The steps for calculating the manual intervention rate in S3 include:

[0083] Obtain the total number of decision-making intervention operations performed by managers within a preset time window from the system log;

[0084] From the system log, obtain the total number of scheduling instructions issued by the system within the same time window;

[0085] The manual intervention rate is calculated by dividing the total number of decision intervention operations by the total number of scheduling instructions;

[0086] In this embodiment, the calculation process of the human intervention rate in step S3 is described; this process is intended to directly and objectively quantify the degree of human managers' veto of AI system decisions;

[0087] The calculation process is performed within a preset time window. , for example, periodic execution within the past 24 hours; the internal logic is that the system accesses and parses its operation logs to accurately count the total number of times within that time window that managers, through PC control platforms or mobile applications, have explicitly revoked, modified, or forced manual control of scheduling instructions automatically issued by the system. At the same time, the system counts the number of people in the same time window from the log. The total number of dispatching instructions issued by the dispatching system to all smart gates, water pumps and other actuators within ;

[0088] According to the above statistical results, the manual intervention rate Calculated by the following formula:

[0089] ;

[0090] because and are all unambiguous count values, so It is a dimensionless ratio between 0 and 1, which can intuitively reflect the intensity of decision-making conflict; The higher the value, the lower the manager's trust in the system's decisions;

[0091] This implementation provides an extremely simple and effective method for quantifying human-machine decision conflicts; it does not rely on any subjective questionnaires or complex behavioral models, but directly utilizes the undeniable log data generated by the system itself, ensuring the objectivity and accuracy of the quantitative results; the human intervention rate As a direct representation of the conflict at the decision-making level, the distortion index at the data level The two complement each other and provide another key dimension of high-credibility input for building a comprehensive human-machine system trust entropy model;

[0092] In view of the defect of existing technology ignoring the conflict between human and machine decision-making, this solution designs a manual intervention rate ,in, : Number of interventions, Total number of instructions; this metric is directly derived from immutable system logs. Every manual intervention is considered a veto of the machine's decision by a human expert, providing an objective and direct quantitative basis for measuring conflict at the decision-making level.

[0093] Time Window and The choice of should match the response cycle of the irrigation scheduling system and the intervention habits of managers. A preferred approach is to analyze historical data to select a duration that most significantly reflects the impact of a single irrigation event and the management decision cycle. For example, this can be set between 6 and 48 hours, and dynamically optimized based on system performance.

[0094] Example 4

[0095] S4 specifically includes:

[0096] The global average distortion index and the human intervention rate are weighted using the weight coefficients determined by the expert calibration method and the historical data regression method. The weighted results are summed to generate the trust entropy of the human-machine system.

[0097] In this embodiment, the trust entropy of the human-machine system in step S4 The solution process of is elaborated in detail; the core of this step is to combine the two indicators of data layer distortion and decision layer conflict into a single trust entropy that can guide the control strategy;

[0098] Trust entropy of human-machine system Defined as the global average distortion index and manual intervention rate The weighted sum of:

[0099] ;

[0100] In this formula and are preset dimensionless weight coefficients used to adjust the relative importance of data distortion and human intervention in the comprehensive trust assessment. The determination method of these two weight coefficients is not arbitrary, but based on a rigorous method to ensure their rationality. A preferred implementation method is to combine expert calibration method and historical data regression method. : Trust entropy of human-machine system; : global average distortion index; : Manual intervention rate;

[0101] Expert calibration method refers to the process of inviting several experienced irrigation district managers to subjectively rate the trust impact of a series of virtual or historical scenarios with varying degrees of data distortion and human intervention at the initial stage of system deployment. By statistically analyzing these rating data, it is possible to preliminarily determine the trust impact of the system. and proportional relationship;

[0102] The historical data regression method refers to the process of using machine learning models such as logistic regression to analyze the historical data after the system has accumulated a certain amount of operating data, using typical events of system trustworthiness and system untrustworthiness marked by experts or operation and maintenance personnel as training labels. and The data is trained to reversely solve the problem of best distinguishing the two states. and value;

[0103] This problem can be formulated as a binary classification task, where the historical The data points are used as feature vectors, and the credible and uncredible events marked by experts are used as labels, which are recorded as 0 and 1 respectively; the optimization objective function aims to find a set of weights , so that The calculated trust entropy can maximize the separation of the two types of sample points, for example, by maximizing the trust entropy of the two types of samples The interval of the distribution on the axis; in addition to logistic regression, other advanced classification algorithms such as support vector machines can also be used to solve the optimal weight coefficient;

[0104] This approach enables weight settings to be adaptively learned and optimized from historical experience;

[0105] This implementation successfully integrates two heterogeneous indicators from different levels, data distortion and decision conflict, into a unified trust metric - human-machine system trust entropy through a simple weighted sum model; more importantly, it provides a set of scientific and implementable weight coefficients. and The determination method avoids the subjectivity and arbitrariness of parameter setting, making the calculation of trust entropy not only reasonable in terms of model but also verified and optimized in terms of parameters, greatly improving the accuracy and credibility of trust assessment results;

[0106] This solution builds trust entropy of human-machine system ,in, : distortion weight, : intervention weight, : average distortion, : The intervention rate, a core indicator, combines the uncertainty of the data level with the uncertainty of the Distrust at the level of representation and decision-making, The representations are inherently fused; the practical significance of this formula is that it projects the health of the system from two dimensions into a single, measurable scalar Above: Weight coefficient and It is not set out of thin air, but determined jointly by expert calibration method and historical data regression method, which ensures the trust entropy The calculation model not only has a reasonable structure, but its parameters also contain historical experience and expert knowledge, and have a high degree of reality fitting ability and credibility.

[0107] Example 5

[0108] S5 specifically includes:

[0109] When the trust entropy of the human-machine system is less than the trust entropy safety threshold, the system operates in the global optimal scheduling mode;

[0110] When the trust entropy of the human-machine system is greater than or equal to the trust entropy safety threshold and less than the trust entropy danger threshold, the system switches to the local enhanced decision-making mode;

[0111] When the trust entropy of the human-machine system is greater than or equal to the trust entropy danger threshold, the system switches to human-machine co-driving mode;

[0112] In this embodiment, the adaptive scheduling strategy switching mechanism of step S5 is described; this mechanism is the outlet for the human-machine trust assessment results to be finally converted into actual control actions, and is the core link for achieving system robustness;

[0113] The system presets two key thresholds: trust entropy security threshold and trust entropy danger threshold The setting of these two thresholds is directly related to business risks, and the values ​​can be determined by rigorous methods, for example, by analyzing trust entropy The historical data distribution of the historical data is set as the 80% quantile and the 95% quantile of the historical data respectively. and or, through risk simulation, the risk that may cause the simulated crop yield reduction rate to exceed a certain tolerance level. The values ​​are set to the corresponding thresholds;

[0114] The system will calculate the trust entropy in real time The system continuously compares these two thresholds and automatically and smoothly switches between the following three preset modes based on the comparison results:

[0115] when When the system is running in the global optimal scheduling mode, this state indicates high system trust, reliable data, and harmonious human-machine collaboration. The digital twin model at the center takes full control of the system, and the optimal scheduling instructions calculated by complex algorithms are executed with the primary goal of optimizing the global water utilization coefficient of the irrigation area or maximizing irrigation benefits.

[0116] when When the system switches to local enhanced decision-making mode; this state is in the warning zone, indicating that the system trust has decreased; the system will actively shrink its control strategy, switching from pursuing global optimization to executing local conservative scheduling based on local historical data and safety rule sets; for example, scheduling instructions will no longer be coordinated across regions, but each sub-region will make independent decisions; at the same time, the system will issue clear system risk warnings to managers, prompting them to pay attention;

[0117] when When the system switches to human-machine co-pilot mode; this state is in the danger zone, indicating that the system trust has seriously deteriorated; the system will trigger the circuit breaker mechanism and actively transfer the core decision-making power to human managers. The function of the AI ​​model will change from an autonomous decision-making subject to an auxiliary decision-making unit, only providing auxiliary decision-making information such as data visualization charts and potential risk point analysis, and no longer automatically issuing control instructions;

[0118] This implementation establishes a clear, hierarchical risk response framework; it binds abstract trust levels to specific control modes, achieving a dynamic balance between system autonomy and reliability. This graceful degradation mechanism avoids catastrophic consequences caused by continuing to execute erroneous global optimization instructions when the system is in a poor state, and also prevents the entire intelligent system from being vetoed and abandoned due to a single point of trust issue. It ensures that no matter what trust state the system is in, there is always an optimal human-computer collaboration mode in operation, greatly enhancing the system's resilience and practicality in the real world.

[0119] Example 6

[0120] In the human-machine co-driving mode, it also includes:

[0121] The system marks associated data sources that cause trust entropy to exceed the limit, guiding operations and maintenance personnel to conduct physical inspections and repairs. When the trust entropy of the human-machine system falls below the trust entropy safety threshold due to repairs or scheduling mode adjustments, the system returns to the globally optimal scheduling mode, forming a negative feedback loop.

[0122] In this embodiment, the human-machine co-pilot mode is enhanced, and a closed-loop mechanism for the system to recover from high-risk states is described;

[0123] In the system After entering the human-machine co-driving mode, in addition to transferring decision-making power, the system will automatically execute a traceability analysis program; the program will identify the current trust entropy The factor with the largest contribution to the value;

[0124] Since the calculation formula of the trust entropy of the human-machine system is the weighted sum of various factors, that is, ,in, is the total number of plots in the irrigation area. Therefore, the traceability analysis can be implemented by: the system calculates and sorts the weighted distortion items of all plots separately The value of , and the weighted intervention By comparing the sizes of these items, the system can directly locate which plot’s sensor is continuously distorted, or whether it is due to excessive human intervention, which has become the cause of the increase in overall trust entropy. The dominant factor of the problem, thus providing accurate troubleshooting guidance for operation and maintenance personnel;

[0125] The system will locate those distortion indices Specific sensors with persistently abnormally high values, or specific scheduling decision events with frequent manual intervention; these identified related data sources and decision events are automatically marked as high-priority troubleshooting items by the system and pushed to relevant personnel through the operation and maintenance management interface, greatly shortening fault location time and guiding operation and maintenance personnel to conduct accurate physical world surveys and system repairs;

[0126] As an extension of this mechanism, this method builds a complete negative feedback closed loop; when the operation and maintenance personnel complete the physical repair according to the system guidance, the relevant sensor data distortion index At the same time, as the system switches to a lower-risk scheduling mode, unreasonable AI decisions are reduced and the manual intervention rate is reduced. will also decrease accordingly; the improvement of these two factors will jointly promote the trust entropy of the human-machine system The value of Falling back to the trust entropy security threshold Below, and after a period of stability, the system will automatically try to recover to the global optimal scheduling mode;

[0127] This implementation elevates the present invention from a passive risk avoidance system to an active, self-healing, and resilient system. By intelligently marking risk sources in high-risk mode, it seamlessly guides users to the key links in problem resolution, achieving a shift from human-machine conflict to human-machine collaborative repair. The establishment of a negative feedback closed-loop mechanism ensures that the system can automatically recover to its optimal working state after experiencing disturbances and trust crises, forming a long-term stable and self-improving intelligent ecosystem, significantly improving the system's vitality and user stickiness.

[0128] Based on the quantitative results of the trust entropy of the human-machine system, this solution realizes the adaptive switching of the scheduling strategy, which is an essential leap compared with the rigid control strategy of the existing technology. The existing technology often operates in an optimal mode. Once the prerequisites, such as data quality, are no longer met, the system performance will drop sharply. In this solution, the trust entropy safety threshold Trust entropy danger threshold The setting is based on the statistical analysis of historical data distribution or the simulation of business risks, which together define a smooth transition range from trust to doubt and then to danger; the system is based on The system switches between the global optimal scheduling mode, the local enhanced decision-making mode, and the human-machine co-driving mode based on the comparison results of the value with the two thresholds. This mechanism enables the system's autonomy level to dynamically match its real-time credibility, pursuing the global optimum when the system is in good condition, actively shrinking the decision-making scope and reducing risks when the credibility decreases, and returning the core decision-making power to humans when the credibility is seriously insufficient, avoiding catastrophic decisions driven by erroneous data and greatly improving the system's resilience and safety in real complex environments.

[0129] Example 7

[0130] Irrigation area water condition monitoring equipment using sensor technology, including:

[0131] processor;

[0132] a memory storing a computer program;

[0133] This embodiment provides an irrigation scheduling device for executing the above method; the device is generally a server or a dedicated industrial computer deployed in the support layer of the irrigation district information management system platform; the core hardware structure includes at least one processor and a memory;

[0134] The processor can be a computing unit such as a central processing unit, a graphics processing unit, an application-specific integrated circuit or a field programmable gate array, and its function is to provide computing power support for executing complex model operations and logical judgments; the memory can be a non-volatile storage medium such as a random access memory, a read-only memory or a hard disk;

[0135] A computer program is fixed or loaded in the memory; the computer program includes a series of instructions that, when read and executed by the processor, can implement all or part of the steps of any of the above methods; for example, the program includes a module for receiving federated learning parameter updates, an algorithm module for calculating a distortion index and an intervention rate, a module for calculating trust entropy, and a logic control module for executing mode switching;

[0136] This embodiment provides a clear physical carrier for the aforementioned complex method; by solidifying the method into a computer program executable by a processor and storing it in a device, the technical solution can be easily deployed on standard server hardware, thereby realizing the transformation from abstract algorithms to specific industrial products, and providing the necessary foundation for the large-scale application and promotion of this technology.

[0137] Example 8

[0138] Irrigation area water condition monitoring system using sensor technology, including:

[0139] The perception and execution layer is composed of sensors and intelligent control actuators deployed in the irrigation area;

[0140] The network transmission layer is responsible for transmitting data and instructions between the perception execution layer and the platform support layer;

[0141] Platform support layer;

[0142] This embodiment provides a complete adaptive irrigation scheduling system that implements the above method. The system is a typical IoT application architecture consisting of three logically and physically cooperating layers.

[0143] The perception and execution layer is the interface between the system and the physical world. It is composed of a large number of devices deployed at the irrigation district site, including various sensors for collecting data and intelligent control actuators for executing control instructions.

[0144] The network transport layer forms the communication backbone of the system. It is responsible for establishing a reliable two-way communication link between the perception and execution layer and the platform support layer. It can use a variety of communication technologies such as cellular networks, LoRa, NB-IoT, or optical fiber. Its core function is to upload sensor data collected by the perception and execution layer to the platform support layer and issue scheduling instructions generated by the platform support layer to the corresponding intelligent control actuators.

[0145] The platform support layer serves as the computing and decision-making core of the system. It is typically composed of a server cluster or cloud platform deployed in a data center, equipped with irrigation scheduling equipment. The platform support layer is responsible for executing all steps of the irrigation area water regime monitoring method using sensor technology, including running the digital twin model, aggregating and processing data, calculating the trust entropy of the human-computer system, making scheduling decisions, and executing policy switching.

[0146] These three layers work together to form a complete closed loop of data perception-transmission-analysis-decision-execution, thus achieving intelligent and adaptive monitoring and scheduling of the entire irrigation district;

[0147] This embodiment integrates the method and equipment into a complete, operational solution by defining a clear three-tier system architecture. This architecture clarifies the functional positioning and interrelationships of different components in the system, and offers good modularity and scalability. It builds a complete cyber-physical system that spans the physical world, transitions to the digital world, and returns to the physical world. This provides system-level guarantees for achieving all the functions and technical effects of the aforementioned method, resulting in a technologically advanced, rationally structured, and implementable intelligent irrigation solution.

[0148] This solution builds a complete negative feedback closed loop through fault tracing and guided repair functions in the human-machine co-pilot mode. When the system enters the human-machine co-pilot mode, it not only passively gives up control, but also actively marks the associated data sources that cause the trust entropy to exceed the limit, thereby guiding the operation and maintenance personnel to conduct accurate physical inspections and repairs. This process directly reduces the global average distortion index. and manual intervention rate , which in turn promotes the trust entropy of human-machine systems fall back; when Once the system recovers below the safety threshold, it can automatically revert to the globally optimal scheduling mode. This design enables the entire system to evolve from a static decision-making tool into a resilient organism that can co-evolve with the environment and humans and possess a certain degree of self-repair ability, ensuring its vitality and sustainable adoption in long-term operation. This is also an advanced feature that existing technical solutions do not have at all.

[0149] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for monitoring water conditions in an irrigation area using sensor technology, characterized in that: The steps include: S1. Receive model parameter updates uploaded by each irrigation sub-region under the federated learning framework, and iteratively update the digital twin model of the center-side irrigation area using the model parameter updates. S2. Obtain sensor measured values ​​and manual intervention data from system logs; S3. Calculate the global average distortion index and human intervention rate. The global average distortion index is used to characterize the deviation between the digital twin model and the physical world reflected by the sensor measurements. The human intervention rate is used to characterize the degree of decision conflict based on human intervention data statistics. S4. Based on the preset weights, combined with the global average distortion index and the manual intervention rate, the trust entropy of the human-machine system is calculated; S5. Compare the trust entropy of the human-machine system with the preset trust entropy safety threshold and trust entropy danger threshold, and adaptively switch the irrigation scheduling mode based on the comparison result; The steps for calculating the global average distortion index in S3 specifically include: Based on the water balance principle and meteorological monitoring data, the theoretical change in soil volume water content of the plot within the time window is calculated; Calculate a normalization factor that only considers the contribution of total water inflow; combine the theoretical change with the actual change in water content measured by the sensor, and use the normalization factor to calculate the sensor data distortion index for each plot; perform spatial smoothing and averaging of the sensor data distortion index for all plots to generate a global average distortion index; The steps for calculating the manual intervention rate in S3 include: Obtain the total number of decision-making intervention operations performed by managers within a preset time window from the system log; From the system log, obtain the total number of scheduling instructions issued by the system within the same time window; The manual intervention rate is calculated by dividing the total number of decision intervention operations by the total number of scheduling instructions; S4 specifically includes: The global average distortion index and the human intervention rate are weighted using the weight coefficients determined by the expert calibration method and the historical data regression method. The weighted results are summed to generate the trust entropy of the human-machine system. S5 specifically includes: When the trust entropy of the human-machine system is less than the trust entropy safety threshold, the system operates in the global optimal scheduling mode; When the trust entropy of the human-machine system is greater than or equal to the trust entropy safety threshold and less than the trust entropy danger threshold, the system switches to the local enhanced decision-making mode; When the trust entropy of the human-machine system is greater than or equal to the trust entropy danger threshold, the system switches to human-machine co-driving mode.

2. The irrigation area water regime monitoring method using sensor technology according to claim 1, characterized in that: In the human-machine co-driving mode, it also includes: The system marks the associated data sources that cause the trust entropy to exceed the limit to guide operation and maintenance personnel to conduct physical inspections and repairs; when the trust entropy of the human-machine system falls below the trust entropy safety threshold due to repairs or scheduling mode adjustments, the system returns to the global optimal scheduling mode, forming a negative feedback closed loop.

3. An irrigation area water regime monitoring device using sensor technology, applied to an irrigation area water regime monitoring method using sensor technology according to any one of claims 1 to 2, characterized in that: include: processor; A memory storing a computer program.

4. An irrigation area water regime monitoring system using sensor technology, applied to the irrigation area water regime monitoring method using sensor technology according to any one of claims 1 to 2, characterized in that: include: The perception and execution layer is composed of sensors and intelligent control actuators deployed in the irrigation area; The network transmission layer is responsible for transmitting data and instructions between the perception execution layer and the platform support layer; Platform support layer.

Images