An intelligent water level monitoring method and device based on double sensors and segmented reporting

Abstract

The application discloses an intelligent water level monitoring method and device based on double sensors and segmented reporting, and belongs to the technical field of intelligent monitoring. By introducing a simplified hydraulics relationship model for physical consistency verification and generating a mechanism confidence factor, the fusion process can be deeply embedded in the hydraulics mechanism, ensuring the physical reasonableness of the data. The fusion weight is dynamically adjusted in combination with the instantaneous reading quality of the sensor, historical reliability data and health index, improving the accuracy and reliability of the high-confidence water level value. In addition, the associated data and mechanism confidence factor are combined to form a multi-physical field coupling feature vector, which enhances the ability to identify the physical nature of complex hydrological events, providing a more accurate and comprehensive data basis for subsequent event semantic analysis. This enables earlier and more accurate identification of potential hydrological anomalies, thereby improving the forward-looking and reliability of monitoring and early warning.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring technology, and in particular to an intelligent water level monitoring method and device based on dual sensors and segmented reporting. Background Technology

[0002] In the construction of smart water conservancy systems, water level monitoring is a core component. Traditional solutions generally employ dual-sensor redundancy configurations and fixed threshold-triggered reporting mechanisms. This approach has significant limitations at the perception level: the fusion processing of multi-source sensor data relies solely on basic numerical verification, such as simple threshold comparisons or arithmetic averages of readings from two sensors, failing to deeply embed hydraulic mechanisms into the data verification process. In complex hydrological scenarios, such as sudden changes in river flow velocity, fluctuations in sediment concentration, or tidal influences, a dynamic coupling relationship exists between water level and physical quantities such as flow velocity and turbidity. Existing technologies lack the ability to dynamically assess this physical consistency in real time, leading to insufficient reliability of heterogeneous data fusion results. This makes it difficult to accurately capture the early physical characteristics and evolution patterns of events such as floods and dam breaks, easily resulting in false alarms or missed alarms. Summary of the Invention

[0003] This application provides an intelligent water level monitoring method and device based on dual sensors and segmented reporting, which can reduce the risk of false alarms or missed alarms. The technical solution is as follows: On the one hand, a smart water level monitoring method based on dual sensors and segmented reporting is provided, the method comprising: Dynamic trust fusion under hydraulic mechanism constraints is performed on the heterogeneous water level raw data collected by dual-body sensors and the associated data collected by multi-physics associated sensors to obtain a high-confidence water level value and a multi-physics coupled feature vector. The dynamic trust fusion is based on the joint judgment of the sensor health evolution trend and simplified hydraulic consistency verification. The event semantic evolution analysis is performed on the multiphysics field coupled feature vector and the historical hydrological event sequence to obtain the semantic type, confidence level and semantic value density of the current hydrological event. The semantic value density is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost. The historical hydrological event sequence is derived from recent high-value density data fragments cached locally. Based on the semantic value density, the high confidence level value, and the current full-link resource status, a segmented reporting strategy package is generated. The segmented reporting strategy package includes a joint scheduling instruction for differentiated sampling frequency configuration, channel parameter selection, and data compression intensity. The segmented reporting strategy package is executed and strategy execution performance data is collected. The performance data is associated and mapped with the semantic type to generate a double-loop evolutionary feedback signal. The double-loop evolutionary feedback signal is used to iteratively update the weight generation rules of the dynamic trust fusion and the calculation parameters of the semantic value density.

[0004] On the one hand, an intelligent water level monitoring device based on dual sensors and segmented reporting is provided, the device comprising: The fusion module is used to perform dynamic trust fusion of heterogeneous raw water level data collected by dual-body sensors and associated data collected by multi-physics associated sensors under the constraints of hydraulic mechanism, so as to obtain a high-confidence water level value and multi-physics coupled feature vector. This dynamic trust fusion is based on the joint judgment of sensor health evolution trend and simplified hydraulic consistency verification. The analysis module is used to perform event semantic evolution analysis on the multi-physics coupled feature vector and the historical hydrological event sequence to obtain the semantic type, confidence level and semantic value density of the current hydrological event. The semantic value density is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost. The historical hydrological event sequence is derived from the recent high-value density data fragments cached locally. The generation module is used to generate a segmented reporting strategy package based on the semantic value density, the high confidence level value and the current full-link resource status. The segmented reporting strategy package contains a joint scheduling instruction for differentiated sampling frequency configuration, channel parameter selection and data compression intensity. The execution module is used to execute the segmented reporting strategy package and collect strategy execution performance data. It then associates and maps the performance data with the semantic type to generate a double-loop evolutionary feedback signal. This double-loop evolutionary feedback signal is used to iteratively update the weight generation rules of the dynamic trust fusion and the calculation parameters of the semantic value density.

[0005] The technical solution provided in this application addresses the lack of deep physical verification in traditional methods during data fusion. By introducing a simplified hydraulic relationship model for physical consistency verification and generating mechanistic confidence factors, the fusion process can be deeply embedded with hydraulic mechanisms, ensuring the physical rationality of the data. Dynamically adjusting the fusion weights by combining instantaneous sensor reading quality, historical reliability data, and health indicators improves the accuracy and reliability of high-confidence water level values. Furthermore, combining associated data with mechanistic confidence factors to form a multi-physics coupled feature vector significantly enhances the ability to identify the physical essence of complex hydrological events, providing a more accurate and comprehensive data foundation for subsequent event semantic analysis. This enables earlier and more accurate identification of potential hydrological anomalies, thereby improving the foresight and reliability of monitoring and early warning systems. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a flowchart of an intelligent water level monitoring method based on dual sensors and segmented reporting provided in an embodiment of this application; Figure 2 This is a flowchart of another intelligent water level monitoring method based on dual sensors and segmented reporting provided in this application embodiment. Detailed Implementation

[0008] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0009] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0010] Dual-body sensors: These refer to two main sensors arranged in a predetermined geometric layout on the monitoring section, used to collect heterogeneous raw water level data. These sensors may employ different measurement principles or have different characteristics to provide redundant and complementary information, thereby improving the reliability of water level measurements.

[0011] Multiphysics-associated sensors: These are sensors used to collect data on other physical quantities related to water level changes, such as flow velocity, turbidity, temperature, or conductivity. This associated data provides additional contextual information for understanding the physical nature of hydrological events.

[0012] Hydraulic mechanism constraints refer to the fundamental physical laws and principles used to describe the movement and behavior of water bodies. In data processing, these mechanisms are used as verification rules to ensure that data on different physical quantities conform to objective physical relationships.

[0013] Dynamic Trust Fusion: This refers to a data fusion technology whose core lies in dynamically adjusting the trust weights of each data source based on factors such as the real-time status of sensors, data quality, and consistency with physical mechanisms, thereby generating more reliable and accurate fusion results.

[0014] High-reliability water level value: refers to the water level measurement result obtained after dynamic trust fusion processing. This value integrates information from multiple data sources and is verified by physical mechanisms and sensor health, thus possessing higher accuracy and reliability.

[0015] Multiphysics coupling eigenvector: refers to a comprehensive data representation that combines and correlates high-confidence water level values ​​with associated data and mechanism confidence factors to comprehensively characterize the current multiphysics state of the water body and its interactions.

[0016] Sensor health evolution trend: This refers to the changing patterns of a sensor's performance indicators (such as accuracy, stability, and drift) during long-term operation. By analyzing this trend, the current operating status and future reliability of the sensor can be assessed, thereby affecting the weight of its data in the fusion process.

[0017] Simplified hydraulic consistency verification: This refers to using a simplified hydraulic relationship model to physically and logically compare and verify the collected raw water level data and associated data in order to determine whether there are contradictions or anomalies between the data, thereby generating a mechanism confidence factor.

[0018] Historical hydrological event sequences refer to a collection of recent hydrological event data fragments with high value density cached locally by the device. This sequence contains information such as the semantic type, evolution process, related data, and policy execution performance of past hydrological events, providing crucial historical context for the analysis of current events.

[0019] Event semantic evolution analysis refers to the in-depth analysis of multiphysics coupled feature vectors and historical hydrological event sequences to identify the type, confidence level, and potential evolution trend of current hydrological events. This analysis aims to transform raw data into "hydrological event semantics" with clear decision-making guidance.

[0020] Semantic type: refers to the labels used to classify and mark hydrological events, such as "normal water level", "slight rise", "peak formation", or "receding stage". It provides a high-level, easy-to-understand description of hydrological phenomena.

[0021] Confidence level: refers to the degree of certainty that the device identifies the semantic type of the current hydrological event. The higher the confidence level, the more reliable the device's judgment of the semantic type.

[0022] Semantic value density is a quantitative indicator that measures the importance and urgency of a hydrological event. It is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost required to transmit data related to the event, reflecting the balance between the potential impact of the event and the resource input.

[0023] Decision Importance Weight: This refers to a weighting coefficient that is preset or dynamically adjusted based on the semantic type of a hydrological event. This coefficient reflects the importance of different types of events to decision-making; for example, flood warning events typically have a higher weight than normal water level reports.

[0024] Resource consumption cost: refers to the estimated resource overhead incurred in transmitting data related to the current hydrological event, including but not limited to communication bandwidth, terminal energy, and processing capacity.

[0025] Segmented reporting strategy package: This refers to a set of scheduling instructions used to guide the data acquisition, processing, and transmission process. This strategy package dynamically configures differentiated sampling frequencies, channel parameter selection, and data compression intensity based on the semantic value density of events and device resource status.

[0026] Differentiated sampling frequency configuration: This refers to dynamically adjusting the frequency of sensor data acquisition based on the semantic value density and urgency of hydrological events. Important or urgent events may trigger a higher sampling frequency to obtain more detailed data.

[0027] Channel parameter selection: refers to dynamically selecting the optimal communication channel, transmission power, or modulation method based on the current network conditions and data priority to ensure the efficiency and reliability of data transmission.

[0028] Data compression strength refers to the degree to which data is compressed before transmission. Higher compression strength can reduce transmission bandwidth requirements, but may increase local processing burden or affect data accuracy.

[0029] End-to-end resource status: refers to the resource availability and load status of all components in the current device (including sensors, communication networks, processing units, etc.). This status is an important constraint for generating segmented reporting strategy packages.

[0030] Policy execution performance data refers to performance metrics collected during or after the execution of segmented policy packages, such as actual energy consumption, network transmission latency, data reporting success rate, and value achievement rate. This data is used to evaluate the effectiveness of the policy.

[0031] Dual-loop evolutionary feedback signal: refers to a multi-level feedback mechanism that uses the association mapping between strategy execution performance data and semantic types to generate signals for iteratively updating device parameters. The "dual loop" typically refers to two feedback loops working collaboratively, one local to the device and one in the cloud.

[0032] Weight generation rules: refer to the algorithms or models used to determine the trust weights of each sensor during the dynamic trust fusion process.

[0033] Calculation parameters: refer to the variables or coefficients used in the calculation of indicators such as semantic value density.

[0034] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0035] In traditional water level monitoring devices, dual-sensor and fixed-threshold reporting mechanisms are widely used. Their data fusion process only performs simple verification operations, failing to embed hydraulic mechanisms into the deep correlation verification of multi-source data. This results in insufficient ability to identify the early physical characteristics of complex hydrological events. The reporting strategy at the decision-making level relies on passive triggering by preset water level thresholds, preventing the transformation of monitoring data into hydrological event semantics with clear decision-making guidance. There is a lack of dynamic correlation between event value and tightly constrained resources such as communication bandwidth and terminal energy. At the device architecture level, data perception, processing, and communication are optimized in isolation, without establishing a joint optimization model for end-to-end resources centered on event value. Furthermore, the lack of a secure and efficient collaborative learning mechanism among edge nodes makes it difficult for the device as a whole to achieve adaptive evolution. Specifically, the lack of physical verification at the perception level leads to decreased data reliability, and rigid resource scheduling at the decision-making level causes a disconnect between device resource allocation and the actual value of the data. This device-level fragmentation further weakens the reliability of monitoring and the foresight of early warning.

[0036] For example, in a water level monitoring device deployed at a river monitoring point in a certain watershed, dual-body sensors and multi-physics associated sensors are used to collect associated data such as water level, flow velocity, and pressure. When encountering continuous heavy rainfall that causes rapid fluctuations in water level, traditional methods only use the difference in readings of the two sensors for threshold verification, failing to verify data consistency based on the hydraulic equations of the river cross-section. For instance, in the rapid flow area of ​​a bend, the sensor readings deviate due to the hydrodynamic characteristics, and the device cannot identify potential risks of overtopping. At the same time, the reporting strategy is fixed at transmitting a complete data packet every 30 minutes. Even if the rate of water level change exceeds the safety threshold, the sampling frequency or compression intensity cannot be dynamically adjusted, resulting in redundant occupation of communication resources or missing data on critical events. In addition, each monitoring node operates independently, and historical high-value hydrological event data is not effectively extracted for optimizing the local decision-making model. The device repeatedly makes misjudgments under similar hydrological conditions, and the lack of a parameter collaborative update mechanism between nodes leads to a continuous weakening of the overall response capability.

[0037] If the above problems are not solved, the monitoring device will be unable to accurately capture the early physical evolution characteristics of hydrological events, which may delay the response to emergencies; the disconnect between resource scheduling and data value will lead to the ineffective consumption of communication bandwidth and terminal energy, reducing the device's ability to operate continuously in harsh environments; the lack of an adaptive evolution mechanism makes it difficult for the device to adapt to the dynamic changes in hydrological conditions, and its reliability will gradually deteriorate during long-term operation, failing to meet the core requirements of smart water conservancy for high reliability and forward-looking early warning.

[0038] Based on this, the technical solution provided in the embodiments of this application is proposed.

[0039] This embodiment provides an intelligent water level monitoring method based on dual sensors and segmented reporting. See [link to relevant documentation]. Figure 1 This includes the following steps.

[0040] 101. Dynamic trust fusion is performed on the heterogeneous raw water level data collected by dual-body sensors and the associated data collected by multi-physics associated sensors under hydraulic mechanism constraints to obtain high-confidence water level values ​​and multi-physics coupled feature vectors. This dynamic trust fusion is based on the joint determination of sensor health evolution trend and simplified hydraulic consistency verification.

[0041] 102. Perform semantic evolution analysis on the multiphysics coupled feature vectors and historical hydrological event sequences to obtain the semantic type, confidence level, and semantic value density of the current hydrological event. The semantic value density is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost. The historical hydrological event sequences originate from recent high-value density data fragments cached locally.

[0042] 103. Based on semantic value density, high confidence level, and current end-to-end resource status, generate a segmented reporting strategy package. This segmented reporting strategy package includes joint scheduling instructions for differentiated sampling frequency configuration, channel parameter selection, and data compression intensity.

[0043] 104. Execute the segmented reporting strategy package and collect strategy execution performance data. Associate and map the performance data with semantic types to generate a double-loop evolutionary feedback signal. This double-loop evolutionary feedback signal is used to iteratively update the weight generation rules and semantic value density calculation parameters of dynamic trust fusion.

[0044] In one implementation, policy execution performance data (such as energy consumption and latency) can be recorded and simply correlated with corresponding semantic types to form historical logs. These logs can be manually analyzed to evaluate the effectiveness of the policy. In another implementation, policy execution performance data can be periodically aggregated and reviewed by a device administrator or expert. Based on the review results, the weight generation rules for dynamic trust fusion or the calculation parameters for semantic value density can be manually adjusted. In yet another implementation, policy execution performance data can be used to directly update some parameters of the local device.

[0045] Finally, the device executes the generated segmented reporting strategy package and simultaneously collects strategy execution performance data. For example, when executing a high-frequency sampling and high-bandwidth transmission strategy, the device records actual energy consumption, data transmission latency, and data reporting success rate. This performance data is then mapped to the semantic type of "moderate flood event" corresponding to this strategy execution. Through this mapping, the device can generate a dual-loop evolutionary feedback signal. For example, if the high-frequency sampling strategy leads to excessive energy consumption or a lower-than-expected data reporting success rate, the feedback signal instructs the device to adjust the dynamic trust fusion weight generation rules (e.g., reducing the trust weight of a power-consuming sensor) or the semantic value density calculation parameters in subsequent decisions (e.g., reassessing the resource consumption cost of a "moderate flood event"). This feedback mechanism enables the device to continuously learn and optimize to adapt to constantly changing environmental and resource conditions.

[0046] This application constructs an intelligent monitoring device that deeply integrates perception, decision-making, and execution. It extracts high-value event semantics from heterogeneous data and dynamically and accurately schedules resources across the entire chain based on this semantics. Under stringent resource constraints, it systematically improves the reliability of monitoring and the foresight of early warning, demonstrating significant technological advancements.

[0047] This application proposes a method for dynamic trust fusion under hydraulic mechanism constraints of heterogeneous raw water level data collected by dual-body sensors and associated data collected by multi-physics associated sensors, to obtain a high-confidence water level value and a multi-physics coupling feature vector. The method includes: determining the initial confidence weight of each sensor based on the instantaneous reading quality and historical reliability data of each sensor; using a simplified hydraulic relationship model to perform physical consistency verification of the heterogeneous raw water level data and the associated data, generating a mechanism confidence factor; determining the target weight of each sensor in the fusion calculation based on the initial confidence weight, the mechanism confidence factor, and the health indicators of each sensor; using the target weight to perform weighted calculation on the heterogeneous raw water level data of the dual-body sensors, which are arranged in a predetermined geometric layout on the monitoring section, to generate the high-confidence water level value; and combining and correlating the associated data with the mechanism confidence factor to obtain the multi-physics coupling feature vector.

[0048] The initial reliability weight for each sensor is determined based on its instantaneous reading quality and historical reliability data. This aims to provide a preliminary weighting benchmark reflecting the sensor's own performance for subsequent fusion calculations. Instantaneous reading quality can be a real-time assessment of the sensor's current output data noise level, volatility, or deviation from adjacent readings. Historical reliability data can be a statistical analysis of the sensor's failure rate, calibration records, maintenance history, or data drift trends during long-term operation.

[0049] A simplified hydraulic relationship model is used to verify the physical consistency between heterogeneous water level raw data and associated data, generating a mechanism confidence factor. Its role is to introduce physical laws to conduct in-depth verification of the data. The simplified hydraulic relationship model can be a simplified mathematical expression based on Manning's formula, continuity equation, or empirical hydraulic formula, used to describe the intrinsic relationship between physical quantities such as water level, flow velocity, flow rate, and turbidity under specific hydrological conditions.

[0050] Based on initial confidence weights, mechanistic confidence factors, and health indicators of each sensor, the target weight of each sensor in the fusion computation is determined. The purpose is to dynamically adjust the contribution of each sensor to more accurately reflect the reliability of the current data. Health indicators can include real-time operating status parameters such as sensor battery level, communication signal strength, internal self-test status, and data transmission packet loss rate.

[0051] The heterogeneous raw water level data from dual-body sensors are weighted using target weights to generate highly reliable water level values. The dual-body sensors are arranged in a predetermined geometric layout on the monitoring section, aiming to improve the accuracy and robustness of water level measurement through multi-source heterogeneous data fusion. The dual-body sensors can be two water level gauges based on different measurement principles, such as an ultrasonic water level gauge and a radar water level gauge. They can be arranged side-by-side, staggered at different heights, or at varying distances on the monitoring section to adapt to different measurement environments and water level ranges.

[0052] By combining and correlating associated data with mechanistic confidence factors, a multi-physics coupled feature vector is obtained. Its role is to integrate physical consistency information into the feature representation of hydrological events, thereby enhancing the event identification capability.

[0053] The above implementation methods effectively address the lack of deep physical verification in traditional data fusion processes. By introducing a simplified hydraulic relationship model for physical consistency verification and generating mechanistic confidence factors, the fusion process can be deeply embedded with hydraulic mechanisms, ensuring the physical rationality of the data. Simultaneously, dynamically adjusting the fusion weights based on sensor instantaneous reading quality, historical reliability data, and health indicators significantly improves the accuracy and reliability of high-confidence water level values. Furthermore, combining associated data with mechanistic confidence factors to form a multi-physics coupled feature vector significantly enhances the ability to identify the physical essence of complex hydrological events, providing a more accurate and comprehensive data foundation for subsequent event semantic analysis. This enables the device to identify potential hydrological anomalies earlier and more accurately, thereby improving the foresight and reliability of monitoring and early warning systems.

[0054] This application further proposes using a simplified hydraulic relationship model to verify the physical consistency between heterogeneous water level raw data and associated data, generating a mechanism confidence factor. This process includes: 201. Determine the current water level change rate based on the historical water level data sequence of the dual-body sensor.

[0055] 202. Based on a simplified hydraulic relationship model, determine the expected range of theoretical associated parameters that match the current rate of change of water level and the current water level value. The theoretical associated parameters shall include at least one of the expected range of flow velocity or the expected range of turbidity.

[0056] 203. Compare the actual associated parameters in the associated data collected by the multi-physics associated sensor with the expected range of the theoretical associated parameters to obtain the parameter deviation.

[0057] 204. Based on the deviation of this parameter, generate a confidence factor for the mechanism that characterizes the physical consistency between the original heterogeneous water level data and the associated data. The greater the deviation of this parameter, the smaller the confidence factor for the mechanism.

[0058] Determining the current rate of water level change aims to dynamically capture the real-time trend of water level changes, providing dynamic input for subsequent theoretical calculations and ensuring the real-time nature and accuracy of physical consistency verification. One approach is to obtain the rate of water level change over time by performing differential calculations or linear regression analysis on historical water level data sequences collected by dual-body sensors within a continuous time window.

[0059] The purpose of determining the expected range of theoretical associated parameters is to use a hydraulic model to predict the reasonable range of other physical parameters related to water level changes based on the current rate of water level change and water level value, so as to serve as a benchmark for verifying actual associated data.

[0060] The parameter deviation is obtained to quantify the degree of difference between the actual associated parameters collected by multiphysics associated sensors and the theoretical expected range, and to provide an objective indicator for evaluating data consistency.

[0061] The generation mechanism confidence factor aims to transform parameter deviation into a quantitative confidence index, reflecting the degree of consistency between heterogeneous water level raw data and associated data in terms of physical mechanism, and providing a basis for subsequent dynamic trust fusion.

[0062] Through the above implementation methods, by capturing water level change trends in real time and combining them with the theoretical range of associated parameters dynamically predicted by the hydraulic model, physical consistency verification no longer relies on static thresholds but can accurately reflect the physical correlation under dynamic hydrological conditions, thus significantly improving the accuracy of physical consistency assessment. Since the mechanistic confidence factor is generated based on dynamic and accurate physical consistency assessment, it can more realistically reflect the strength of physical correlation between heterogeneous data. When this confidence factor is used for dynamic trust fusion, sensor weights can be allocated more rationally, thereby generating more reliable high-confidence water level values. Accurate and reliable high-confidence water level values ​​and multi-physics coupled feature vectors generated based on accurate mechanistic confidence factors provide high-quality input for subsequent semantic analysis of hydrological events. This enables the device to identify the early physical characteristics of complex hydrological events earlier and more accurately, providing a more solid data foundation for early warning and decision-making.

[0063] This application further proposes a step to combine and correlate associated data with mechanistic confidence factors to obtain a multiphysics coupled feature vector, including: dividing the intermediate result of the fusion processing of heterogeneous raw water level data from dual-body sensors into a time-series context window, and extracting the actual feature sequence of the associated data within this time-series context window; generating a theoretical feature sequence corresponding to the actual feature sequence based on the simplified hydraulic relationship model and the high-confidence water level value; weighting the residual sequence between the actual feature sequence and the theoretical feature sequence using the mechanistic confidence factor to obtain a weighted residual feature; and jointly encoding the high-confidence water level value, the actual feature sequence, the theoretical feature sequence, and the weighted residual feature to obtain the multiphysics coupled feature vector.

[0064] Specifically, "dividing the time-series context window based on the intermediate results of the fusion processing of heterogeneous raw water level data from dual-body sensors" refers to using the high-confidence water level value after dynamic trust fusion processing as a reference to determine a time period for focused analysis of associated data. This "intermediate result of fusion processing" typically refers to the high-confidence water level value, which provides reliable information about the current water level status. The time-series context window can be dynamically adjusted based on the changing trend or stability of the high-confidence water level value. For example, when the water level changes rapidly, a shorter window can be used to capture instantaneous changes. When the water level changes gradually, a longer window can be used to accumulate richer contextual information. Alternatively, a fixed-length time-series window can be preset, such as dividing a window every few minutes or hours, and using the high-confidence water level value within that window as a representative reference for that window.

[0065] "Extracting the actual feature sequence of the associated data within the specified time frame" refers to extracting features reflecting dynamic changes from data collected by multi-physics associated sensors within the defined time window. This associated data may include flow velocity, turbidity, temperature, etc. Extracting the actual feature sequence can be achieved by directly sampling the original associated data to form a time series. Alternatively, it can involve statistical analysis of the associated data, such as calculating the average, maximum, minimum, standard deviation, or rate of change within the window, thereby forming a multi-dimensional feature vector sequence. Furthermore, time-frequency analysis methods, such as wavelet transform or Fourier transform, can be used to extract the characteristic components of the associated data at different frequencies to more comprehensively characterize its dynamic properties.

[0066] "Generating a theoretical characteristic sequence corresponding to the actual characteristic sequence based on the simplified hydraulic relationship model and the high-confidence water level value" refers to using known physical laws and reliable water level information to predict the expected performance of associated data under current water level conditions. The "simplified hydraulic relationship model" can be an empirical or semi-empirical formula such as the Manning formula or the Chezy formula, used to describe the relationship between water level and associated parameters such as flow velocity and flow rate. Generating the theoretical characteristic sequence can be achieved by substituting the high-confidence water level value into the simplified hydraulic model, calculating the theoretically expected associated parameter values ​​in real time, and arranging them in chronological order. Alternatively, it can be achieved by obtaining the corresponding theoretical associated parameter sequence from a pre-established lookup table or empirical curve based on the high-confidence water level value.

[0067] "Weighted residual features are obtained by weighting the residual sequences between the actual and theoretical characteristic sequences using the mechanistic confidence factor" refers to correcting the differences between actual observations and theoretical predictions by introducing a mechanistic confidence factor to highlight the parts with high physical consistency. This "mechanistic confidence factor" is an indicator that measures the physical consistency between heterogeneous raw water level data and associated data; a higher value indicates better consistency. The generation of weighted residual features can be achieved by directly applying the mechanistic confidence factor as a multiplicative weight to the difference sequence between the actual and theoretical characteristic sequences. Alternatively, it can be achieved by adjusting the residual sequence using a nonlinear function (such as a sigmoid or exponential function) based on the magnitude of the mechanistic confidence factor, thereby reducing the impact of residuals when physical consistency is poor and preserving or amplifying the true physical meaning of residuals when physical consistency is high.

[0068] "Jointly encoding the high-confidence water level value, the actual feature sequence, the theoretical feature sequence, and the weighted residual feature to obtain the multiphysics coupled feature vector" refers to integrating information from multiple sources and types into a unified and representative data structure. Joint encoding can be achieved by directly concatenating these features to form a high-dimensional vector, for example, combining the high-confidence water level value, the statistics of the actual feature sequence (such as mean and variance), the statistics of the theoretical feature sequence, and the statistics of the weighted residual feature. Alternatively, machine learning or deep learning methods, such as the encoder part of an autoencoder or recurrent neural network, can be used to input these features, allowing the network to learn and output a more compact and information-rich low-dimensional feature vector.

[0069] In one specific implementation, at the monitoring section, dual-body sensors continuously collect heterogeneous raw water level data, which, along with accompanying data collected by multi-physics sensors (e.g., a flow velocity sensor and a turbidity sensor), are input to an edge computing unit. The edge computing unit first performs a dynamic trust fusion process to obtain a high-confidence water level value. Then, based on the current high-confidence water level value, the unit defines a shorter 30-second time-series context window if the water level value has changed more than a preset threshold in the past 5 minutes, or a longer 2-minute time-series context window if the water level value changes steadily. Within the defined time-series context window, the edge computing unit extracts actual feature sequences from the flow velocity and turbidity sensors, for example, sampling once per second to obtain raw time series of flow velocity and turbidity. Simultaneously, the edge computing unit uses a pre-configured simplified hydraulic relationship model (e.g., an empirical model based on Manning's formula), combined with the current high-confidence water level value, to calculate the theoretically expected flow velocity and turbidity values ​​at that water level and generates the corresponding theoretical feature sequences. Next, the device uses the mechanistic confidence factor generated during the dynamic trust fusion process to weight the residuals between the actual and theoretical flow velocity sequences, as well as the residuals between the actual and theoretical turbidity sequences. For example, a higher mechanistic confidence factor (indicating good physical consistency) results in a larger residual weight. Conversely, a lower mechanistic confidence factor results in a smaller residual weight, and can even suppress residuals at low confidence levels. Finally, the edge computing unit jointly encodes the current high-confidence water level value, the extracted actual flow velocity and turbidity feature sequences, the generated theoretical flow velocity and turbidity feature sequences, and the weighted flow velocity and turbidity residual features using a pre-trained neural network encoder, outputting a fixed-dimensional multiphysics coupled feature vector.

[0070] The above implementation method effectively addresses the potential issue of insufficient physical consistency when combining associated data and mechanistic confidence factors. By introducing high-confidence water level values ​​as the basis for dividing the time-series context window, the starting point for data analysis is ensured to have high reliability. Extracting actual feature sequences within the time-series context window accurately captures the dynamic changes in associated data. More importantly, generating theoretical feature sequences based on a simplified hydraulic relationship model and high-confidence water level values ​​provides a solid physical benchmark for actual observations. Weighting the residual sequences using mechanistic confidence factors effectively filters out noise and uncertainty in the final weighted residual features, highlighting true physical deviations and ensuring that the feature vector accurately reflects the hydraulic mechanism. Finally, jointly encoding high-confidence water level values, actual feature sequences, theoretical feature sequences, and weighted residual features results in a multi-physics coupled feature vector that is not only comprehensive in information but also possesses high physical consistency, significantly improving the ability to identify the essential characteristics of complex hydrological events and providing more accurate and reliable input for subsequent semantic analysis of hydrological events.

[0071] This application further proposes a method for performing semantic evolution analysis on multiphysics coupled feature vectors and historical hydrological event sequences to obtain the semantic type, confidence level, and semantic value density of the current hydrological event. This method includes: performing feature analysis on the multiphysics coupled feature vectors and jointly parsing the historical hydrological event sequences to extract trend features characterizing the evolution of hydrological events; matching and reasoning the trend features with a predefined hydrological event semantic rule base to obtain the semantic type and confidence level of the current hydrological event, wherein the matching and reasoning process incorporates temporal context verification based on historical hydrological event sequences; and determining the semantic value density based on the semantic type, confidence level, and estimated resource consumption cost.

[0072] The process involves feature analysis of multiphysics coupling feature vectors and joint analysis of historical hydrological event sequences to extract trend features characterizing the evolution of hydrological events. The aim is to identify and quantify the dynamic development patterns or potential evolution directions of hydrological events from these features and historical event sequences. The extraction of trend features allows the device to gain insights into the dynamic evolution of events from static data snapshots, thereby identifying the physical nature of complex hydrological events earlier and more accurately. One approach is to use a deep learning-based time series prediction model, such as using a gated recurrent unit (GRU) network to model the sequence of multiphysics coupling feature vectors and combining it with event labels and timestamps from historical hydrological event sequences. By learning the contextual dependencies of events, trend features characterizing their evolution can be generated. Another approach is to use a graph neural network (GNN)-based method to construct an event graph from historical hydrological event sequences. Nodes represent event states, edges represent state transitions, and multiphysics coupling feature vectors are used as attributes of nodes or edges. Graph convolution operations are then used to extract the topological and temporal trend features of event evolution.

[0073] The process involves matching and reasoning trend features against a predefined hydrological event semantic rule base to obtain the semantic type and confidence level of the current hydrological event. This matching and reasoning process incorporates temporal context verification based on historical hydrological event sequences. The aim is to compare the extracted trend features with the predefined hydrological event semantic rules to identify the category of the current hydrological event and assess the reliability of the identification. Simultaneously, historical event sequences are used to verify the temporal rationality of the reasoning results. By introducing temporal context verification, misjudgments caused by instantaneous data fluctuations or local anomalies can be effectively avoided, improving the accuracy and robustness of event identification. One implementation approach is to construct a fuzzy logic-based reasoning device. Trend features are used as fuzzy input, and fuzzy reasoning is performed through a fuzzy rule base (e.g., defining "rapid water level rise" and "abnormal flow velocity" as a combination of "early flood" events) to obtain the membership degree of each semantic type, which is then converted into a confidence level. Temporal context verification can be performed by checking whether the currently inferred event type has a reasonable temporal correlation with event types that have occurred in historical event sequences (e.g., whether it conforms to the evolutionary chain of "rainfall-water level rise-flood"). Another approach is to use case-based reasoning (CBR), which uses trend features as query cases, searches for similar historical cases in historical hydrological event sequences, and performs inference based on the semantic type and evolutionary path of these similar cases. Temporal context verification can then adjust the confidence level of the inference results by evaluating the temporal matching degree between the current case and historical cases, as well as the consistency of the event evolution path.

[0074] Based on semantic type, confidence level, and estimated resource consumption cost, semantic value density is determined to quantify the comprehensive value of current hydrological events. This value considers not only the severity of the event itself and the reliability of its identification but also the resource costs required for data transmission and processing. By incorporating resource consumption cost, intelligent optimization allocation of limited device resources can be achieved, ensuring that high-value event data is processed and reported preferentially and efficiently. One implementation approach is to design a multi-objective optimization model, where semantic type corresponds to different decision importance weights, confidence level serves as a weight adjustment factor, and estimated resource consumption cost is considered as a cost item. Semantic value density can be defined as the optimization result that maximizes the product of event importance and confidence level while minimizing resource consumption under given resource constraints. This model can be solved using linear programming or nonlinear programming algorithms. Another implementation approach is to employ a hierarchical multi-criteria decision-making method, treating semantic type, confidence level, and estimated resource consumption cost as three independent evaluation dimensions. By assigning weights to each dimension (e.g., expert scoring or machine learning training), and then weighted summation or product form, the comprehensive semantic value density is calculated.

[0075] Through the above implementation methods, raw, heterogeneous water level data and associated data can be deeply fused and analyzed to transform hydrological event semantics with clear business implications and decision-making value. This scheme, by jointly analyzing multi-physics coupled feature vectors and historical hydrological event sequences, can accurately capture the dynamic evolution patterns of hydrological events, overcoming the limitations of traditional methods that rely solely on static threshold judgments, and significantly improving the ability to identify the early physical characteristics of complex hydrological events. Simultaneously, the integration of temporal context verification during the matching inference process effectively enhances the accuracy and robustness of event identification, avoiding false alarms and missed alarms. More importantly, this scheme combines the semantic type and confidence level of events with the estimated resource consumption cost, dynamically quantifying the semantic value density of events. This allows the device to intelligently and precisely manage and schedule limited communication bandwidth and terminal energy resources, focusing on event value, thereby systematically improving the reliability of water level monitoring and the foresight of early warnings under stringent resource constraints.

[0076] This application further proposes a method for jointly analyzing multiphysics coupling feature vectors and historical hydrological event sequences to extract trend features characterizing the evolution of hydrological events. The steps include: constructing an event state transition probability matrix based on historical hydrological event sequences; performing dynamic time warping matching between the multiphysics coupling feature vectors and the historical feature vectors corresponding to the historical hydrological event sequences to obtain the similarity and phase difference between the current event and the historical event evolution stages; and determining the trend probability vectors of the current hydrological event evolving into various potential subsequent event types based on the event state transition probability matrix, similarity, and phase difference, and using these trend probability vectors as trend features.

[0077] Specifically, the event state transition probability matrix is ​​a mathematical model used to quantify the probabilistic relationships between different hydrological event types over time. It analyzes historical data to statistically determine the likelihood of transitioning from one hydrological event state to another, thus providing a statistical basis for predicting future event trends. This matrix can be constructed using a Markov chain model, which discretizes the historical hydrological event sequence into a series of states, then counts the number of transitions from any state i to any state j, and divides this by the total number of transitions from state i to obtain the transition probability P(j|i). These probabilities are organized into a matrix, which is the event state transition probability matrix. Alternatively, more complex probabilistic graphical models such as Hidden Markov Models (HMMs) or Conditional Random Fields (CRFs) can be used. Training algorithms can learn the state transition probabilities and observation probabilities from historical event sequences to construct a more refined event state transition probability matrix.

[0078] Dynamic Time Warping (DTW) is an algorithm used to measure the similarity between two time series, even when they are non-linearly stretched or compressed along the time axis. Here, it compares the current multiphysics coupled eigenvector with the historical eigenvectors corresponding to each historical event in a historical hydrological event sequence. This overcomes the common velocity inconsistency problem in time series data, accurately quantifying the similarity (similarity) between the current and historical events during their evolution and their alignment difference (phase difference) along the time axis. A classic DTW algorithm can be used, constructing a distance matrix and finding a path with the minimum cumulative distance to align the two time series. The cumulative distance along the path serves as a measure of similarity, while the path itself reveals the alignment of the time series, allowing the calculation of the phase difference. Alternatively, improved DTW algorithms, such as FastDTW or ConstrainedDTW, can be used. These algorithms provide a time alignment path while calculating similarity, thus deriving the phase difference.

[0079] A trend probability vector quantifies the probability distribution of a current hydrological event evolving into various potential subsequent event types. It integrates the statistical regularities of historical event evolution (event state transition probability matrix), the similarity between the current event and historical events, and their alignment in temporal evolution (similarity and phase difference), thus providing a more comprehensive and accurate prediction of future trends. Using this vector as a trend feature can provide rich dynamic evolution information for subsequent event semantic analysis. The calculated similarity can be used as weights to weight and correct the row vectors in the event state transition probability matrix related to the current event state, while the phase difference is used to adjust the transition probabilities temporally. Finally, through normalization, a vector representing the probability distribution of future event types is obtained. Alternatively, a fusion model can be designed, such as a model based on Bayesian networks or neural networks, using the event state transition probability matrix, similarity, and phase difference as input features. By learning the complex nonlinear relationships in historical data, it can directly output the trend probability vector of the current hydrological event evolving into various potential subsequent event types.

[0080] As a specific implementation method, when performing feature analysis on multiphysics coupled feature vectors and joint analysis on historical hydrological event sequences to extract trend features characterizing the evolution of hydrological events, the device can maintain a historical hydrological event database. This database stores the type, duration, key feature vector sequences, and actual transition records between all hydrological events that occurred within a past period. Based on these historical records, a Markov chain model can be constructed to generate an event state transition probability matrix. For example, if historical data shows that the "normal water level" state has an 80% probability of remaining at "normal water level," a 15% probability of transitioning to "slight rise," and a 5% probability of transitioning to "rapid rise," these probability values ​​will be filled into the matrix. This matrix can be updated periodically to adapt to changes in the hydrological environment. When the multiphysics coupled feature vector of the current moment is received, the device performs dynamic time warping (DTW) matching with the feature vector sequences of each historical event stored in the historical hydrological event sequence. For example, the FastDTW algorithm can be used to efficiently calculate the distance between the current feature vector sequence and the historical sequence and find the optimal alignment path. By analyzing this regularized path, the overall time offset or local time stretching / compression of the current event relative to historical events can be calculated, thus obtaining the phase difference. Finally, the device uses the constructed event state transition probability matrix, the calculated similarity, and the phase difference to determine the trend probability vector of the current hydrological event evolving into various potential subsequent event types. Specifically, the transition probabilities related to the current event state in the event state transition probability matrix can first be corrected based on the phase difference. For example, if the phase difference indicates an accelerated evolution of the current event, the probability of transitioning to the "rapid evolution" event type can be appropriately increased, while the probability of transitioning to the "slow evolution" event type can be decreased. Next, the similarity obtained from DTW matching (normalized to a value between 0 and 1) is used as a weighting coefficient and weighted and fused with the transition probabilities corrected for the phase difference. Finally, the fused probability values ​​are normalized to obtain a probability distribution vector representing the evolution of the current hydrological event into potential subsequent event types such as "normal water level," "slight rise," "rapid rise," and "peak flood passage." This probability vector serves as the trend feature for subsequent event semantic analysis.

[0081] Through the above technical solutions, this method can significantly improve the accuracy and predictive ability of capturing the evolution patterns of hydrological events. First, by constructing an event state transition probability matrix based on historical hydrological event sequences, this method provides a solid statistical foundation for predicting event evolution trends, ensuring that trend characteristics reflect the inherent regularity of hydrological events. Second, the introduction of a dynamic time warping matching mechanism effectively solves the common nonlinear time alignment problem in time series data, enabling precise quantification of the similarity and phase difference between the current event and historical events during their evolution stages. This overcomes the limitations of traditional methods in handling time series distortions, resulting in a more refined and accurate capture of event evolution trends. Finally, by comprehensively utilizing the event state transition probability matrix, similarity, and phase difference, this method can determine the trend probability vector of the current hydrological event evolving into various potential subsequent event types. This vector comprehensively integrates historical statistical patterns and real-time matching information. This makes the generated trend features not only statistically significant but also dynamically real-time, providing more reliable and richer input for subsequent semantic evolution analysis of events. This greatly improves the accuracy of assessing the semantic type, confidence level, and semantic value density of current hydrological events, thereby enhancing the early warning foresight and decision support capabilities of intelligent water level monitoring devices.

[0082] This application further proposes a method for determining the trend probability vectors of the evolution of the current hydrological event into various potential subsequent event types based on the event state transition probability matrix, similarity, and phase difference. The method includes: adjusting the probability weights of corresponding transition paths in the event state transition probability matrix based on the phase difference to obtain a time-corrected transition probability matrix; using similarity as a weighting coefficient and weighting it with each transition probability starting from the current state in the time-corrected transition probability matrix to obtain the initial trend probability vectors for each potential subsequent event type; and performing a probability distribution transformation based on a normalized exponential function on the initial trend probability vectors to obtain the final trend probability vector.

[0083] The phase difference refers to the time or state offset between the current stage of a hydrological event and the corresponding stage in a historical hydrological event sequence. This phase difference can be positive (indicating the current event lags behind historical events) or negative (indicating the current event precedes historical events), and its magnitude reflects the degree of temporal deviation. The event state transition probability matrix is ​​a two-dimensional matrix where rows and columns represent different hydrological event states, and each element represents the probability of transitioning from one state to another. This matrix is ​​typically obtained through statistical analysis or Markov chain modeling of historical hydrological event sequences and is used to describe the inherent laws governing the evolution of hydrological event states. Adjusting the probability weights of corresponding transition paths in the event state transition probability matrix refers to dynamically correcting the probability values ​​of specific state transition paths in the matrix based on the phase difference between the current hydrological event and historical events. For example, when the phase difference indicates a rapid evolution of the current event, the probability weight for transitioning to a "higher level" or "more urgent" event state can be appropriately increased, while the probability weight for transitioning to a "stable" or "declining" event state can be decreased. Specific adjustment methods can include: a linear decay or growth function based on the phase difference, where the probability weights change linearly with the absolute value of the phase difference; or a method based on a Gaussian kernel function, where the closer the phase difference is to zero, the smaller the adjustment magnitude, and the greater the deviation, the larger the adjustment magnitude. Through these adjustments, the transition probability matrix can be made to better reflect the real-time temporal characteristics of the current hydrological event, thus obtaining a time-corrected transition probability matrix.

[0084] Similarity refers to the degree of matching or similarity between the current hydrological event and matching events in the historical hydrological event sequence in terms of their evolutionary stages. This similarity is typically a value between 0 and 1, where 1 represents a perfect match and 0 represents a complete mismatch. Similarity can be calculated using Dynamic Time Warping (DTW) or other sequence matching algorithms. Using similarity as a weighting coefficient means that during the fusion process, the similarity value directly affects the contribution of the time-corrected transition probabilities to the initial trend probability vector. For example, a higher similarity indicates a closer match between the evolutionary patterns of the current event and historical events, resulting in a stronger dominance of the corrected transition probabilities in the fusion process. Weighted fusion involves combining the transition probabilities from the current state in the time-corrected transition probability matrix with the similarity to generate the initial trend probability vector for each potential subsequent event type. For example, a weighted average can be used, weighting the corrected transition probabilities with a baseline probability (such as the average transition probability of all event types), with the weight being the similarity. Alternatively, the corrected transition probabilities can be directly multiplied by the similarity to obtain the initial trend probability for that event type. The initial trend probability vector is a one-dimensional vector in which each element corresponds to a potential subsequent hydrological event type and represents the initial probability of that event type occurring.

[0085] The normalized exponential function (NEF) is a function that transforms any real value into a probability distribution, such as the Softmax function. This function maps each element of the input vector to the (0,1) interval and ensures that the sum of all output elements is 1, thus forming an effective probability distribution. Transforming the initial trend probability vector using the NEF based on the normalized exponential function involves taking the initial trend probability vector as input and processing it through the NEF to obtain the final trend probability vector. This transformation can amplify the differences between larger and smaller values ​​in the initial probability vector, making the probability distribution clearer, while avoiding the problems of non-standardized probability values ​​or unbalanced distributions that may result from weighted fusion. The final trend probability vector is a normalized probability distribution, where each element represents the final probability of the current hydrological event evolving into the corresponding potential subsequent event type.

[0086] The following example illustrates this. Assume the current hydrological event is in the "rapid water level rise" phase. By performing dynamic time warping matching with historical hydrological event sequences, the similarity between the current event and the historical event's evolution phase is found to be 0.85, with a phase difference of +2 time steps (indicating the current event lags behind the historical event by 2 time steps). First, based on this phase difference, the pre-constructed event state transition probability matrix is ​​adjusted. For example, if the original matrix shows a transition probability of 0.6 from "rapid water level rise" to "torrential flood peak," 0.3 from "sustained high water level," and 0.1 from "water level receding," since the phase difference is positive, indicating the event evolution may be slower or longer-lasting, the transition probability to "torrential flood peak" can be appropriately reduced, while the transition probability to "sustained high water level" can be increased. Specifically, a linear decay function can be used to adjust the probability of "torrential rain and flood peak" to 0.6*(1-0.05*2)=0.54, and the probability of "sustained high water level" to 0.3*(1+0.05*2)=0.33. Other probabilities are then adjusted and normalized accordingly to obtain a time-corrected transition probability matrix. Next, the similarity of 0.85 is used as a weighting coefficient and weighted with the time-corrected transition probabilities. For example, a baseline probability vector (e.g., [0.4, 0.3, 0.3]) can be set, and the initial trend probability vector can be calculated as: `initial trend probability = similarity * corrected transition probability + (1-similarity) * baseline probability`. For example, for the "torrential rain and flood peak" type, its initial trend probability might be 0.85*0.54+(1-0.85)*0.4=0.459+0.06=0.519. Finally, a normalized exponential function (such as the Softmax function) is applied to the obtained initial trend probability vector to transform it into the final trend probability vector. For example, if the initial trend probability vector is [0.519, 0.325, 0.156], after processing with the Softmax function, the final trend probability vector may be [0.55, 0.30, 0.15], where the sum of all elements is 1, indicating that the probability of the current hydrological event evolving into "rainstorm peak", "sustained high water level", and "water level receding" is 55%, 30%, and 15%, respectively.

[0087] The above-described implementation methods effectively address the issues of temporal bias, insufficient utilization of similarity information, and unreasonable probability distribution in hydrological event evolution prediction. Specifically, adjusting the transition probability weights based on phase difference allows the prediction model to better adapt to the real-time temporal characteristics of the current hydrological event, avoiding prediction errors caused by temporal mismatch. Integrating similarity as a weighting coefficient into the probability fusion process fully utilizes the matching degree between the current event and historical event evolution patterns, enhancing the rationality and reliability of the prediction results. Furthermore, transforming the initial trend probability vector using a normalized exponential function ensures the smoothness, stability, and interpretability of the final trend probability vector, avoiding the impact of extreme values ​​or uneven distributions on subsequent decisions. These improvements collectively enhance the accuracy and stability of hydrological event evolution trend prediction, providing a more accurate foundation for event semantic analysis for intelligent water level monitoring devices. This enables more effective support for early warning and resource scheduling, improving the overall performance and foresight of the device.

[0088] This application proposes an improved method for event semantic evolution analysis, comprising: pre-screening candidate inference rule sets from a predefined hydrological event semantic rule base based on trend features; inputting multiphysics coupling feature vectors and trend features into the candidate inference rule sets, performing forward chain inference, and generating an inference result set containing one or more possible semantic types and preliminary confidence levels; performing temporal consistency verification on the inference result set using historical hydrological event sequences, eliminating semantic types that contradict the event evolution temporal logic, and correcting the preliminary confidence levels of the remaining semantic types to obtain the verified results; and determining the semantic type and corresponding confidence level of the current hydrological event based on the corrected confidence levels of each semantic type in the verified results.

[0089] The trend feature refers to the dynamic information characterizing the evolution pattern of hydrological events, obtained through semantic evolution analysis of multiphysics coupled feature vectors and historical hydrological event sequences. It can be a probability vector indicating the likelihood of the current event evolving into different subsequent event types. The predefined hydrological event semantic rule base is a knowledge base storing the definitions, feature patterns, evolution paths, and related reasoning logic of various hydrological events (e.g., floods, droughts, debris flows, water quality anomalies). Each rule in the rule base can include triggering conditions, event types, confidence calculation methods, etc. The pre-screening step of selecting candidate reasoning rule sets aims to quickly identify the subset of rules most relevant to the current hydrological situation and most likely to be triggered from the vast rule base based on the current trend features. This can significantly reduce the computational load of subsequent reasoning and improve reasoning efficiency. For example, if the trend feature strongly points to "flood evolution," rules related to floods can be prioritized, while rules related to drought or water quality anomalies can be ignored. Pre-screening can be based on rule metadata tags, similarity matching between rule conditions and trend features, or through pre-defined decision trees or classifiers. The multiphysics coupling feature vector is a comprehensive feature representation obtained by combining high-confidence water level values ​​obtained through dynamic trust fusion under hydraulic mechanism constraints of heterogeneous raw water level data collected by dual-body sensors and associated data collected by multiphysics associated sensors. It contains detailed physical state information of the current hydrological environment. This forward chain reasoning is a reasoning method that starts from known facts (i.e., the multiphysics coupling feature vector and trend features) and gradually derives conclusions (i.e., semantic types and confidence levels) by applying rules from a rule base. It checks whether the preconditions of the rules are met; if so, it triggers the execution of the rules and generates new facts or conclusions. For example, it can use a reasoning engine based on a production rule system or a reasoning mechanism based on logic programming. The reasoning result set contains one or more possible semantic types and initial confidence levels, meaning that in the initial reasoning stage, the device may identify multiple potential hydrological event types and assign an initial confidence level to each type based on the degree of rule matching. This historical hydrological event sequence refers to recently cached high-value-density data fragments, containing the semantic type, evolution process, duration, and associated sensor data and policy execution performance data of past hydrological events. This temporal consistency check is a crucial verification step designed to ensure that the initial inference results are reasonable in the temporal dimension. The evolution of hydrological events typically follows certain temporal logic; for example, a "flood" is likely to occur after a "heavy rain," but not immediately after a "drought."The verification process compares the semantic types in the inference result set with the event evolution patterns recorded in historical hydrological event sequences, identifying and eliminating semantic types that are illogical in temporal sequence. For example, an event state transition diagram can be constructed or a Hidden Markov Model (HMM) can be used to evaluate the rationality of the event sequence. This operation of eliminating semantic types that contradict the temporal sequence of event evolution directly solves the problem mentioned in the introduction, namely, avoiding the adoption of erroneous semantic types that do not conform to the temporal sequence of event evolution. The initial confidence of the remaining semantic types is then corrected. For semantic types that pass the temporal consistency verification, their initial confidence is adjusted according to their degree of fit with the historical evolution pattern. For example, if a semantic type highly matches the evolution path of a historical event, its confidence may be increased. Conversely, if the fit is low, it may be appropriately reduced. Correction can be achieved using methods such as Bayesian updates, fuzzy logic adjustments, or weight allocation based on expert knowledge. The result after verification is a set containing one or more semantic types and their corrected confidence after temporal consistency verification and confidence correction. The step of determining the semantic type and corresponding confidence level of the current hydrological event typically involves selecting the semantic type with the highest corrected confidence level from the validated results as the final semantic type of the current hydrological event. If multiple semantic types have similar highest confidence levels, the device can make a final decision based on preset priority rules, risk assessment, or further expert device judgment.

[0090] As a specific implementation method, when performing event semantic evolution analysis on multiphysics coupling feature vectors and historical hydrological event sequences, the following steps can be followed: First, based on the trend characteristics of the current hydrological event, for example, if the trend characteristics show that the current water level is rising rapidly and the flow velocity is abnormally accelerating, the device will pre-select rules related to "flood," "waterlogging," and "river overflow" from a predefined hydrological event semantic rule base, while temporarily ignoring rules unrelated to "drought" and "water pollution." This rule base can be stored in local memory and defined in XML or JSON format. Each rule contains a unique ID, triggering conditions (a logical expression based on the multiphysics coupling feature vector and trend characteristics), event type label, and a preliminary confidence calculation formula. Second, the current multiphysics coupling feature vector (e.g., containing high-confidence water level values, flow velocity, turbidity, etc.) and trend characteristics (e.g., a vector indicating a probability of 0.8 for "flood evolution") are input into the pre-selected candidate inference rule set. The device can use a rule engine based on the Rete algorithm to perform forward chain inference. For example, if a rule's conditions are "water level rise rate exceeds a threshold AND trend characteristics indicate flood evolution," and the current data meets these conditions, the rule is triggered, generating a preliminary inference result, such as "semantic type: moderate flood, preliminary confidence level: 0.75." Through inference, multiple such results may be generated, forming an inference result set, for example, containing "moderate flood (0.75)" and "channel blockage (0.6)." Next, the device uses historical hydrological event sequences to perform a temporal consistency check on this inference result set. Historical hydrological event sequences can be stored as structured data of timestamp-event type-confidence level-duration-related features. The device queries historical sequences, for example, finding that "channel blockage" events typically occur when water levels rise slowly and flow rates decrease, and usually do not occur immediately after a rapid rise in water levels. "Moderate flood," on the other hand, is usually a subsequent event of "heavy rain" or "upstream discharge." If the current inference result of "channel blockage" does not conform to historical evolution patterns—for example, in the context of rapidly rising water levels—the device will consider "channel blockage" to be logically contradictory in terms of time sequence and thus remove it from the inference result set. For "moderate flood," the device will assess its consistency with historical flood evolution paths. If the consistency is high, its initial confidence level may be revised from 0.75 to 0.85. If the consistency is moderate, it may be revised to 0.70. Finally, based on the results after time sequence consistency verification and revision, the device will select the semantic type with the highest revised confidence level as the final semantic type of the current hydrological event. For example, if the revised result only shows "moderate flood (0.85)," the device determines the semantic type of the current hydrological event as "moderate flood" with a confidence level of 0.85.The finalized semantic type and confidence level will be used for subsequent semantic value density calculations and the generation of segmented reporting strategy packages.

[0091] This application introduces a trend-feature-based candidate inference rule pre-screening mechanism, which effectively narrows the inference scope and improves inference efficiency. More importantly, by using historical hydrological event sequences to verify the temporal consistency of the preliminary inference results, this method can eliminate semantic types that contradict the logical sequence of event evolution and correct the preliminary confidence of the remaining semantic types. This mechanism significantly enhances the accuracy and reliability of hydrological event semantic type identification, avoiding semantic type temporal logical contradictions that may result from direct inference. Therefore, this method can provide more accurate and reliable semantic types and confidence levels of current hydrological events, providing a solid foundation for subsequent semantic value density calculation and the generation of segmented reporting strategy packages. This enables the entire intelligent water level monitoring device to more accurately understand the hydrological situation and, based on this, to conduct more rational and efficient resource scheduling, ultimately improving the device's early warning foresight and overall operational efficiency.

[0092] This application further proposes a method for determining semantic value density, which is based on semantic type, confidence level, and estimated resource consumption cost. Specifically, the method includes: calculating dynamic event weights using a nonlinear correction function based on preset decision importance weights and confidence levels corresponding to semantic types; estimating the estimated resource consumption cost of transmitting data related to the current hydrological event based on historical resource consumption records in historical hydrological event sequences, as well as semantic types and confidence levels; and inputting the dynamic event weights and estimated resource consumption cost into an online optimizer to solve for the semantic value density. The online optimizer aims to maximize the utility function defined by the dynamic event weights and estimated resource consumption cost within the current decision period.

[0093] Semantic type refers to the category or nature of the current hydrological event, such as "normal water level," "mild flooding," "moderate flooding," and "sudden debris flow warning." Its function is to assign high-level semantics to the event so that the device can understand its potential impact and importance. Possible implementation methods include matching multi-physics coupled feature vectors to specific semantic labels through a predefined rule base; or using machine learning classification models (such as support vector machines or neural networks) to classify the feature vectors and output the corresponding semantic type. Confidence represents the reliability or certainty of the device's judgment on the semantic type of the current hydrological event. High confidence means the device has a high degree of certainty about the judgment result, while low confidence indicates uncertainty. Its function is to quantify the reliability of the semantic judgment and provide a basis for subsequent decision-making. Possible implementation methods include using the probability value output by the classification model; for example, if a softmax layer is used, its output can be directly used as the confidence level. Alternatively, it can be calculated based on the posterior probability or membership degree obtained through Bayesian inference or fuzzy logic devices according to the strength of evidence. Preset decision importance weights refer to pre-defined numerical values ​​for different semantic types, reflecting their importance at the decision-making level. Their function is to assign different decision priorities to different types of events. Possible implementation methods include having hydrological experts or managers manually configure a fixed weight value for each semantic type based on historical experience and risk assessment. Alternatively, the weights for each semantic type can be calculated and updated using a risk assessment model based on the risk level of a specific region, season, or historical event. A nonlinear correction function is a mathematical function used to nonlinearly combine or adjust the preset decision importance weights and confidence levels to generate dynamic event weights. Its function is to avoid the biases that linear combinations may introduce, allowing dynamic event weights to more accurately and precisely reflect the actual value of events. Possible implementation methods include the Sigmoid or Tanh functions, mapping the linear combination results to the range of 0 to 1 (or -1 to 1) and introducing nonlinear characteristics. Alternatively, piecewise linear or polynomial functions can be used, employing different correction curves depending on the confidence level interval to adapt to the weight adjustment needs at different confidence levels. Dynamic event weights combine the pre-defined importance of an event with its current confidence level, aiming to more accurately reflect the actual decision-making value of the current hydrological event and provide a more refined value assessment for resource allocation. Possible implementation methods include applying a nonlinear correction function to the product of the pre-defined decision importance weight and the confidence level, or applying the nonlinear correction function separately to the pre-defined decision importance weight and the confidence level, and then combining them. Historical resource consumption records in historical hydrological event sequences refer to detailed records of the actual computing resources, communication bandwidth, storage space, energy, and other resources consumed by the device when handling similar hydrological events in the past. Their purpose is to provide empirical data for predicting future resource consumption. Possible implementation methods include storing the resource consumption data for each event processing in a local or cloud database, associated with the event type and processing results.Alternatively, resource usage indicators related to specific event processing can be extracted by analyzing the device's operation logs. Estimated resource consumption cost refers to the resource cost the device is expected to incur for transmitting data related to the current hydrological event. Its purpose is to quantify the cost of data transmission so as to weigh value against cost. Possible implementation methods include regression analysis based on historical data, using historical resource consumption records to train a regression model, and predicting resource consumption based on the semantic type and confidence level of the current event. Alternatively, it can be based on rule and parameter models, calculating resource consumption based on parameters such as event type, data volume, transmission distance, and network conditions, combined with a pre-defined resource consumption model. An online optimizer is a real-time running optimization algorithm or module used to solve for the optimal semantic value density within the current decision-making cycle, based on dynamic event weights and estimated resource consumption cost. Its purpose is to dynamically weigh event value against resource cost, ensuring that the device prioritizes the most valuable events under resource constraints. Possible implementation methods include solvers based on linear programming or integer programming, transforming the utility function and constraints into a mathematical programming problem and solving it using the appropriate solver. Alternatively, based on heuristic or metaheuristic algorithms, when the problem complexity is high, algorithms such as genetic algorithms, particle swarm optimization, and simulated annealing can be used to find an approximate optimal solution in a finite time. A utility function is a mathematical expression used to quantify the device's benefit, defined by dynamic event weights and estimated resource consumption costs, within a specific decision-making cycle. Its role is to provide a clear optimization objective for the online optimizer. Possible implementations include simple ratio functions, such as the value derived from unit resource consumption, or more complex weighted combination functions to balance value and cost.

[0094] As a specific implementation method, the method for determining semantic value density can be implemented as follows: First, for the calculation of dynamic event weights, a preset decision importance weight table can be set, for example, the weight of "normal water level" is 0.1, "mild flooding" is 0.5, "moderate flooding" is 0.8, and "sudden debris flow warning" is 1.0. Assume that the semantic type of the current hydrological event is identified as "moderate flooding," and its confidence level is 0.9. At this time, a Sigmoid function can be used as a nonlinear correction function, for example, `f(x)=1 / (1+exp(-k*x))`, where `k` is an adjustment parameter. The dynamic event weight can be calculated as `W_dynamic=f(preset decision importance weight * confidence level)`. Specifically, `W_dynamic=1 / (1+exp(-5*(0.8*0.9)))=1 / (1+exp(-3.6))` yields a dynamic event weight value between 0 and 1, which non-linearly adjusts the actual value of the event as the confidence level changes. Secondly, regarding the estimation of resource consumption costs, the device can maintain a historical resource consumption database, recording the average energy consumption, transmission time, and data volume required to transmit data for different semantic types of events at different confidence levels. When a "moderate flood" event (confidence level 0.9) is identified, the device can retrieve all historical resource consumption records for "moderate flood" events from the database. Then, a machine learning-based regression model, such as a gradient boosting tree model, can be used. This model takes semantic type, confidence level, and data volume as input features and outputs estimated energy consumption (e.g., joules), transmission bandwidth (e.g., KBps), and storage space (e.g., KB). For example, the model might estimate that transmitting data on the current "moderate flooding" event would require 500 joules of energy and 10 KB of bandwidth. Finally, the calculated dynamic event weight (e.g., 0.97) and the estimated resource consumption cost (e.g., 500 joules of energy, 10 KB of bandwidth) are input into an online optimizer. This online optimizer can be a convex optimization solver based on the Lagrange multiplier method or KKT conditions. The utility function can be defined as `U=W_dynamic-λ_energy*C_energy-λ_bandwidth*C_bandwidth`, where `λ_energy` and `λ_bandwidth` are preset resource penalty coefficients, and `C_energy` and `C_bandwidth` are the estimated energy and bandwidth consumption, respectively. The optimizer solves this utility function in real time to maximize the total utility within the current decision period, thus obtaining the final semantic value density. For example, if the optimizer's solution is 0.85, this value is the semantic value density of the current event, which comprehensively reflects the importance of the event and the transmission cost.

[0095] Through the above implementation methods, the actual value and transmission cost of hydrological events can be accurately quantified, thus solving the problem of accurately linking event value and resource consumption in resource allocation. Specifically, by introducing a nonlinear correction function to calculate dynamic event weights, this scheme can more realistically and precisely reflect the decision importance of events under different confidence levels, avoiding the biases that may be caused by traditional linear assessments. Simultaneously, the estimated resource consumption cost based on historical data and current event characteristics significantly improves the accuracy of transmission cost estimation, enabling the device to comprehensively consider resource input. Finally, by dynamically balancing event value and resource cost through an online optimizer, this scheme ensures that the calculation of semantic value density can reflect resource constraints in real time. This allows for the priority identification and processing of the most valuable hydrological events within limited device resources, effectively avoiding the drawbacks of resource allocation being disconnected from actual event value, and significantly improving the decision-making efficiency and resource utilization rate of intelligent water level monitoring devices in complex hydrological environments.

[0096] This application further proposes a method for estimating the estimated resource consumption cost of transmitting data related to current hydrological events based on historical resource consumption records, semantic types, and confidence levels in historical hydrological event sequences. The method includes: selecting historical events with the same or similar semantic types from the historical hydrological event sequences based on semantic type, and extracting the corresponding historical resource consumption records to obtain a relevant historical consumption dataset; determining a redundancy adjustment coefficient and an expected number of transmission retransmissions based on confidence levels; and inputting the relevant historical consumption dataset, redundancy adjustment coefficient, and expected number of transmission retransmissions into a resource consumption prediction model to calculate the estimated resource consumption cost.

[0097] The semantic type refers to the classification description of the nature, severity, or potential impact of the current hydrological event, such as "minor flood," "moderate drought," or "heavy rain and waterlogging." It is typically obtained through semantic evolution analysis of multiphysics coupling feature vectors and historical hydrological event sequences. During the screening process, "same or similar" can refer to completely identical semantic types, or events belonging to the same major category or possessing similar characteristics within a predefined hydrological event classification system. For example, the degree of "similarity" between semantic types can be quantified using semantic similarity algorithms (such as those based on word vectors or ontology knowledge graphs), or by using pre-defined event association rules. The historical hydrological event sequence is a data set accumulated during the long-term operation of the device, bearing timestamps and event semantic tags. It records past hydrological events and their related information. Extracting historical resource consumption records corresponding to historical events refers to obtaining the actual amount of resources consumed at the time of transmitting data related to that event, such as energy consumption (joules), communication bandwidth usage (bits / second), transmission delay (milliseconds), or processing time (milliseconds). The relevant historical consumption dataset is a collection of data consisting of these selected historical events and their corresponding resource consumption records.

[0098] This confidence level is an indicator measuring the reliability of the semantic type judgment of the current hydrological event. It is typically a value between 0 and 1, with higher values ​​indicating more reliable judgment. Based on this confidence level, the redundancy adjustment coefficient and the expected number of retransmissions can be determined for cost estimation. The redundancy adjustment coefficient adjusts the redundancy of data transmission. For example, when the confidence level is low, more redundant data packets may need to be sent or stronger error correction coding may be used to improve the success rate of data transmission, which increases resource consumption. The expected number of retransmissions refers to the average number of times a data packet is expected to be retransmitted under the current network and event conditions. Lower confidence levels generally mean greater uncertainty in the event judgment, which may lead to an underestimation or overestimation of the urgency or importance of data transmission, thus requiring a higher retransmission expectation to ensure data delivery, which also increases resource consumption. These coefficients and expected values ​​can be dynamically adjusted according to the confidence level using a preset mapping function or lookup table.

[0099] This resource consumption prediction model is a mathematical model or algorithm that can predict resource consumption based on input parameters. The model can take various forms, such as statistical regression models, machine learning-based predictive models (e.g., support vector machines, neural networks), or rule-based expert devices. Its inputs include relevant historical consumption datasets (providing baseline and trend information), redundancy adjustment coefficients (reflecting additional consumption due to increased uncertainty), and expected transmission retransmission counts (reflecting additional consumption due to increased transmission reliability requirements). By comprehensively analyzing and calculating these inputs, the model can output a quantified estimated resource consumption cost, which can be a single numerical value (e.g., total energy consumption) or a multi-dimensional vector (e.g., a combination of energy, bandwidth, and latency).

[0100] In one specific implementation, when the device detects that the semantic type of the current hydrological event is "moderate flood warning," it first filters out all related historical events with semantic types such as "moderate flood warning" or "severe rainstorm" from the locally cached historical hydrological event sequences. Then, the device extracts historical resource consumption records, such as average energy, average bandwidth, and average latency consumed by data transmission during these historical events, forming a historical consumption dataset related to "moderate flood warning." Assuming the current confidence level for "moderate flood warning" is 0.75, the device converts this confidence level into a redundancy adjustment coefficient (e.g., 1.2, indicating a need for 20% redundancy) and an expected number of retransmissions (e.g., 0.3 times), according to a preset mapping rule. Finally, the device inputs this historical consumption dataset, the redundancy adjustment coefficient of 1.2, and the expected number of retransmissions of 0.3 times into a pre-trained resource consumption prediction model based on support vector regression. The model integrates these inputs, calculates and outputs an estimated resource consumption cost. For example, it is estimated that transmitting data related to this "moderate flood warning" will consume 150 joules of energy, occupy 50 kbps of bandwidth, and produce an average latency of 200 milliseconds.

[0101] Through the above implementation methods, the resource consumption cost of transmitting data related to current hydrological events can be more accurately estimated based on semantic type and confidence level. This accurate estimation avoids resource waste or data transmission failures caused by inaccurate estimation, thus ensuring more reliable calculation of semantic value density. Furthermore, this helps the online optimizer to more accurately weigh event value against resource consumption when generating segmented reporting strategy packages, making resource allocation and scheduling more rational and effective, ultimately improving the overall operational efficiency and early warning capabilities of the intelligent water level monitoring device.

[0102] This application proposes a method for generating segmented reporting strategy packages, comprising: generating segmented reporting strategy packages based on semantic value density, high-confidence water level values, and the current end-to-end resource status; the segmented reporting strategy package includes joint scheduling instructions for differentiated sampling frequency configuration, channel parameter selection, and data compression intensity; determining a dynamic priority strategy for resource allocation based on semantic value density and referring to strategy execution performance data recorded in historical hydrological event sequences; constructing a cross-stage resource scheduling optimization function with the objective of maximizing the expected value utility in the next decision cycle, and constrained by the hydrological situation reflected by the high-confidence water level value and the current end-to-end resource status; solving the cross-stage resource scheduling optimization function to obtain optimized scheduling parameters for sensor sampling, data communication, and local processing stages, and generating segmented reporting strategy packages containing execution instructions based on the optimized scheduling parameters.

[0103] Semantic value density is a quantitative indicator that measures the dynamic balance between the importance of a specific hydrological event in the current context and the resource consumption required to obtain information about that event. It comprehensively considers the decision-making importance weight of the event and the resource consumption cost, aiming to identify events that are both critical and have high information acquisition efficiency. A possible implementation method is to dynamically adjust the value density calculation through a pre-defined event type-value mapping table combined with a real-time resource consumption model. Alternatively, a reinforcement learning-based mechanism can be used to continuously optimize the value density calculation through trial and error and feedback, adapting to constantly changing environmental and resource conditions. The strategy execution performance data recorded in historical hydrological event sequences refers to the actual execution effect of strategies (such as sampling frequency, channel selection, and compression intensity) adopted by the device when handling similar hydrological events in the past, including but not limited to indicators such as data reporting success rate, energy consumption, and transmission latency. This data is used to evaluate and optimize the formulation of future strategies. A possible implementation method is to store detailed logs of each strategy execution (including strategy parameters, actual resource consumption, data transmission results, etc.) in a local or cloud database and perform periodic analysis. Alternatively, a time-series analysis-based model can be built to extract trends and patterns in strategy effectiveness from historical data, enabling more accurate predictions of future strategy performance. Dynamic prioritization strategies for resource allocation are mechanisms that adjust the order and proportion of resource allocation in real time based on the semantic value density of events, historical performance data, and the current device state. Their purpose is to ensure that high-value events receive priority access to the necessary computing, communication, and storage resources. A possible implementation could be a multi-level queue scheduler that assigns tasks of different priorities to different processing queues, allocating more processing time slices and bandwidth to higher-priority queues. Alternatively, a bidding-based resource allocation model could be used, where each event "bids" for resources based on its dynamic priority, and devices allocate resources according to the bids.

[0104] Maximizing the expected value utility in the next decision-making cycle is the core objective of constructing the optimization function. This aims to maximize the total value the device can acquire in the next time period through current resource scheduling decisions. The value utility here considers not only the semantic value of the data itself but also factors such as the probability of successful data transmission and timeliness. A possible implementation is to define the expected value utility as a weighted sum function, where the weights are determined by the semantic value density and data transmission success rate, while considering value decay due to data timeliness. Alternatively, a Markov decision process-based framework can be used to model the expected value utility as the cumulative expected value of future rewards, maximizing this expected value through optimization strategies. High-reliability water level values ​​reflect the water situation by dynamically trusting and fusing high-precision, high-reliability water level data obtained through the heterogeneous water level raw data from dual-body sensors with associated data from multi-physics sensors. This data, combined with its changing trends and historical data, reveals the overall condition and potential risks of the current aquatic environment. Its role is to provide crucial physical environmental context information for resource scheduling, such as determining whether a flood warning or dry season is in effect. Possible implementation methods include comparing high-confidence water level values ​​with preset water level thresholds to determine whether the current water situation falls into different levels such as normal, warning, or dangerous, and combining this with the water level change rate to assess the urgency of the water situation. Alternatively, machine learning models can be used to analyze the time series of high-confidence water level values ​​to identify specific water situation patterns (such as rapid rise, slow fall, or stability) and use them as a representation of the water situation. The current end-to-end resource status refers to the real-time status of all available resources along the entire data transmission and processing link from the sensor end to the cloud, including but not limited to the remaining power of sensor nodes, the available bandwidth of communication modules, network latency, local processor computing power, and storage space. Its role is to provide actual available resource constraints for resource scheduling. Possible implementation methods include periodically collecting the operating parameters of each hardware module (such as the battery management unit, communication chip, and CPU) and summarizing them in a central resource manager for unified management. Alternatively, a distributed resource monitoring agent can be used to monitor resource usage in real time on each node and report the status information to the scheduling center through a lightweight protocol. The cross-link resource scheduling optimization function is a mathematical model that aims to comprehensively consider factors such as semantic value density, water situation, and overall resource status to collaboratively optimize resource allocation across different links, including sensor sampling, data communication, and local processing. Its goal is to find the optimal combination of scheduling parameters to maximize expected value utility while satisfying various resource constraints. A possible implementation could be a mixed-integer linear programming model, where decision variables include sampling frequency, channel selection, and compression ratio, the objective function is expected value utility, and constraints include energy, bandwidth, and latency.Alternatively, it could be an optimization model based on heuristic algorithms (such as genetic algorithms or particle swarm optimization), suitable for solving complex scheduling problems that are nonlinear and nonconvex.

[0105] Solving the cross-stage resource scheduling optimization function involves using appropriate algorithms and computational methods to find the solution from the constructed optimization function that satisfies all constraints and maximizes or minimizes the objective function. Its role is to transform the abstract optimization model into concrete scheduling parameters. Possible implementation methods include using commercial or open-source optimization solvers (such as Gurobi, CPLEX, PuLP) to solve linear programming or mixed-integer programming problems. Alternatively, for more complex nonlinear or combinatorial optimization problems, metaheuristic algorithms (such as simulated annealing, ant colony optimization) or approximation algorithms can be used to find suboptimal solutions. The optimization scheduling parameters for sensor sampling, data communication, and local processing stages are specific values ​​or configurations obtained after solving the optimization function, used to guide the operation of each stage. For example, parameters for the sampling stage could be sampling frequency or sampling interval. Parameters for the data communication stage could be channel selection, transmission power, and modulation scheme. Parameters for the local processing stage could be data compression rate, preprocessing algorithm selection, etc. Its role is to transform the optimization results into executable instructions. A possible implementation method is to store these parameters in a configuration table for each module to query and apply at runtime. Alternatively, these parameters can be directly encoded into control commands and sent to the corresponding hardware or software modules via bus or network. The segmented reporting strategy package is a collection of specific execution instructions generated based on optimized scheduling parameters. It defines detailed operating procedures for how sensors sample, how data is communicated, and how local processing occurs within a specific time period. Its role is to transform optimization decisions into an operable, phased sequence of instructions. A possible implementation could be an XML or JSON configuration file containing parameters such as sampling frequency, communication protocol, and compression algorithm for different time periods (or different event types). Alternatively, it could be a script file containing sequentially executed hardware control commands and software processing flows.

[0106] The above implementation methods effectively address the problems of traditional water level monitoring devices, such as the disconnect between resource allocation and event value, the inability to maximize decision-making effectiveness under stringent resource constraints, low resource utilization efficiency, and delayed device response. Specifically, by introducing a dynamic prioritization strategy, the device can intelligently adjust the resource allocation order based on the actual importance of hydrological events and their historical performance, ensuring that high-value events receive priority. Simultaneously, a cross-link resource scheduling optimization function is constructed, taking into account the hydrological situation reflected by high-reliability water level values ​​and the current overall resource status, making resource scheduling decisions more scientific and precise, and adaptable to real-time changes in the environment and resource constraints. This not only significantly improves the data reporting efficiency and reliability of the device under limited resources but also ensures that key hydrological events can be identified and reported in a timely and accurate manner, thereby enhancing the foresight of early warnings and the effectiveness of decision-making.

[0107] This application proposes a method for determining a dynamic priority strategy for resource allocation. This method is based on semantic value density and references strategy execution performance data recorded in historical hydrological event sequences. Specifically, it includes: mapping the semantic value density to a preset initial strategy priority level; performing performance feedback correction on the initial strategy priority level based on strategy execution performance data associated with historical events of the same or similar semantic type in the historical hydrological event sequence to obtain a corrected priority level; and calibrating the corrected priority level based on the current end-to-end resource status to obtain a dynamic priority strategy, which is used to guide resource allocation.

[0108] Semantic value density is an indicator that measures the dynamic balance between the importance of a current hydrological event and the resource cost required for its data transmission. It comprehensively considers the decision-making importance weight of the event and the resource consumption cost, reflecting the urgency and value of reporting the event data under the current device status. When determining the dynamic prioritization strategy for resource allocation, semantic value density serves as the primary value input, providing a foundation for subsequent priority evaluation. It can be obtained by the online optimizer based on dynamic event weights and estimated resource consumption costs, or by looking up a pre-defined lookup table based on semantic type, confidence level, and estimated resource consumption cost. Historical hydrological event sequences are a collection of recently cached high-value-density data fragments, containing the semantic type, confidence level, semantic value density, and corresponding strategy execution performance data of past hydrological events. This sequence provides valuable historical experience for the device, used to analyze the evolution patterns of hydrological events and evaluate the actual effects of different strategies. When determining the dynamic prioritization strategy for resource allocation, historical hydrological event sequences can be used to provide historical data similar to current events for performance feedback correction, or to construct event state transition models to predict future trends. Strategy execution performance data refers to indicators such as actual energy consumption, network transmission latency, event data reporting success rate, and value achievement rate collected during the execution of segmented reporting strategy packages. This data reflects the performance of a specific reporting strategy in actual operation. Strategy execution performance data can be stored on local devices or uploaded to the cloud for centralized management and analysis. It can serve as a basis for evaluating the quality of strategies or as training samples for machine learning models.

[0109] The initial strategy priority level is a resource allocation priority initially determined based on semantic value density. It transforms the abstract value index of semantic value density into discrete or continuous priority values ​​that the device can recognize and manipulate. The initial strategy priority level can directly map the semantic value density to different priority intervals according to a preset mapping function (e.g., piecewise linear function, exponential function), or it can be obtained by consulting a predefined priority mapping table. Performance feedback correction is a process of adjusting the initial priority based on historical strategy execution results. It uses strategy execution performance data associated with historical events of the same or similar semantic type in historical hydrological event sequences to evaluate the rationality of the initial priority setting and make corresponding adjustments. Performance feedback correction can be performed using rule-based expert devices; for example, if historical data shows that a certain priority has a low reporting success rate under specific conditions, its priority is lowered. Alternatively, machine learning-based methods can be used to train a model to predict the deviation of actual performance under a given initial priority and make corrections accordingly. The corrected priority level is the priority after performance feedback correction; it incorporates lessons learned from historical strategy execution on top of the initial priority. This priority level more accurately reflects resource allocation needs after considering historical performance than the initial priority. The revised priority level can be a single value or a priority range, providing a more reliable input for subsequent resource status calibration. The current end-to-end resource status refers to all available resource information possessed by the device at the current moment, including but not limited to the remaining energy of terminal devices, available communication bandwidth, local processor load, and storage space. These resource states are dynamically changing and directly impact the actual execution effect of the strategy. The current end-to-end resource status can be obtained through real-time monitoring by sensors or through statistics and reporting by the device's internal resource management module. Calibration refers to further adjusting the revised priority level based on the current end-to-end resource status to ensure that the resource allocation strategy matches the current actual resource availability. The calibration process considers the device's current resource constraints, avoids allocating priorities beyond available resources, and optimizes resource utilization. Calibration can employ threshold-based adjustments; for example, when available bandwidth falls below a certain threshold, all priorities are downgraded. Alternatively, it can use optimization algorithm-based methods to maximize overall device utility while satisfying resource constraints. The dynamic priority strategy is the final resource allocation priority scheme determined after multiple levels of revision and calibration. It integrates the semantic value of events, the performance of historical strategy execution, and the current resource status of the device, enabling it to guide the device in resource allocation in real time and adaptively. Dynamic priority strategies can be represented as a set of priority rules or a priority scheduling table, directly influencing the generation of scheduling instructions such as sensor sampling frequency, channel parameter selection, and data compression intensity.

[0110] As a specific implementation method, in an intelligent water level monitoring device, after the device receives the high-confidence water level value and multi-physics coupling feature vector output by the dynamic trust fusion module, the event semantic evolution analysis module outputs the semantic type, confidence level, and semantic value density of the current hydrological event. Assume a "moderate flood warning" event is detected, with a semantic value density of 0.85 (range 0-1). First, the device can map this semantic value density of 0.85 to a preset initial strategy priority level. For example, the device can define a mapping rule: a semantic value density between 0.8 and 1.0 corresponds to "high priority," between 0.5 and 0.8 corresponds to "medium priority," and between 0 and 0.5 corresponds to "low priority." Therefore, the "moderate flood warning" event is initially determined to be "high priority." Second, the device will perform performance feedback correction on this "high priority" based on strategy execution performance data recorded in historical hydrological event sequences. The device retrieves historical events with the same or similar semantic types as "moderate flood warning" and analyzes performance metrics such as data reporting success rate, transmission latency, and value achievement rate when these historical events were classified as "high priority." For example, if historical data shows that even "high priority" events only have a data reporting success rate of 70% during certain network congestion periods, far below the expected target of 95%, the device can revise the current "high priority" based on this historical experience. For instance, it can revise it to "high priority - transmission parameters need optimization" or fine-tune its internal values ​​to obtain a revised priority level. This revision can be a rule-based adjustment, such as reducing the priority value by 0.1 if the historical success rate is below a threshold. Alternatively, it can use a small neural network model to predict and output the revised priority value based on historical performance data. Finally, the device calibrates the revised priority level based on the current end-to-end resource status. For example, if the remaining energy of the current terminal device is less than 20%, or the available communication bandwidth is only 30% of normal, the device will further adjust the revised priority based on these real-time resource constraints. Even if an event is corrected to "high priority - transmission parameters need optimization," if current resources are extremely scarce, the device may calibrate it to "medium-high priority" and instruct the use of more aggressive data compression and a lower sampling frequency in the segmented reporting strategy packets to ensure that data can be transmitted rather than being completely lost due to insufficient resources. This calibration process can be a dynamic adjustment based on resource thresholds; for example, when energy falls below a certain percentage, all priority values ​​are uniformly reduced by a fixed value. Alternatively, a linear programming model can be used to reallocate priority weights under current resource constraints. Ultimately, after calibration, the device obtains a dynamic prioritization strategy to guide resource allocation.

[0111] The above implementation significantly improves the accuracy and adaptability of resource allocation in intelligent water level monitoring devices. First, mapping semantic value density to initial strategy priority levels ensures that preliminary resource allocation decisions quickly respond to the intrinsic value of events. Second, introducing a performance feedback correction mechanism based on historical strategy execution performance data allows the device to learn from past experience, avoiding repetitive and inefficient resource allocation patterns, thereby improving the accuracy and reliability of priority setting. Finally, calibrating priorities based on the current end-to-end resource status ensures that the final dynamic priority strategy fully considers real-time resource constraints, avoiding resource waste or under-allocation, and maximizing resource utilization efficiency. This multi-level dynamic adjustment mechanism enables the device to generate more intelligent and efficient segmented reporting strategy packages when facing complex and ever-changing hydrological events and resource environments. This maximizes the timely and reliable reporting of high-value hydrological event data under limited resource conditions, enhancing the overall monitoring device's early warning foresight and decision support capabilities. This application proposes a method for constructing a cross-link resource scheduling optimization function, aiming to maximize the expected value utility in the next decision cycle, and constrained by the water situation reflected by the high-confidence water level value and the current full-link resource status. The method includes: using semantic value density as a value weight coefficient, and constructing the expected value utility as the optimization objective term together with the expected transmission success rate of data packets; determining the transmission urgency level corresponding to the water level range where the high-confidence water level value is located, and quantifying the transmission urgency level as the maximum allowable reporting delay constraint; quantifying the remaining terminal energy and available communication bandwidth in the current full-link resource status as inequality constraints; and constructing a cross-link resource scheduling optimization function by comprehensively optimizing the objective term, the maximum allowable reporting delay constraint, and the inequality constraints. This cross-link resource scheduling optimization function is used to jointly solve for the sampling period, communication channel, and compression algorithm.

[0112] Semantic value density indicates the dynamic ratio between the importance of a current hydrological event to decision-making and the resource consumption required to obtain information about that event. As a value weighting coefficient, it reflects the urgency and potential impact of the event. For example, it can be directly used as a multiplier factor in an optimization function, or mapped using a nonlinear function as a weight. The expected transmission success rate of a data packet indicates the probability that a data packet can be successfully transmitted from the monitoring terminal to the cloud or data center under current network conditions. This success rate can be obtained based on historical transmission data statistics, real-time channel quality assessment, or network topology analysis. For example, it can be estimated by statistically analyzing the data packet loss rate under the same channel conditions over a past period, or assessed in real-time through probe packet transmission and reception acknowledgments. Expected value utility is an indicator that measures the comprehensive value that a device can extract from data under a specific resource scheduling strategy. It combines semantic value density with the expected transmission success rate of data packets to ensure that high-value data is transmitted preferentially and reliably. For example, it can be obtained by simply multiplying the semantic value density by the expected transmission success rate, or by using a more complex utility function with weighted combinations. The optimization objective term is the objective expression that needs to be maximized in the cross-stage resource scheduling optimization function. It represents the best effect that the device hopes to achieve within the current decision-making cycle.

[0113] High-confidence water level values ​​are water level measurements with high confidence obtained after dynamic trust fusion processing. These values ​​reflect the actual water level of the current water body. Water level ranges are pre-defined water level ranges with specific significance based on hydrological characteristics or management needs. For example, they can be divided into "normal water level ranges," "warning water level ranges," and "flood water level ranges." These ranges can be determined based on historical hydrological data statistics or set according to water conservancy engineering design standards. Transmission urgency levels assess the urgency of data reporting based on the water level range in which the high-confidence water level value is located. Higher water levels or more drastic changes generally correspond to higher transmission urgency levels. For example, multiple levels such as "low," "medium," "high," and "urgent" can be set, or continuous numerical values ​​can be used. Maximum allowable reporting delay constraints are obtained by quantifying the transmission urgency levels, representing the maximum allowable time delay from data acquisition to successful reporting. Higher transmission urgency levels result in smaller maximum allowable reporting delay constraints. For example, an "urgent" level can be quantified as a 10-second delay constraint, and a "normal" level as a 5-minute delay constraint.

[0114] The current end-to-end resource status indicates the resource usage and availability along the entire data transmission link from the sensor to the data receiver. The remaining battery power of the terminal indicates the remaining charge of the monitored terminal device's battery. This can be read from the battery management chip or estimated through voltage sampling. Available communication bandwidth indicates the effective bandwidth of the current communication channel available for data transmission. This can be obtained from the network protocol stack or measured in real-time through a channel sounding mechanism. Inequality constraints are mathematical expressions used in the optimization function to limit the range of variable values; they ensure that resource scheduling strategies are feasible under actual physical constraints. For example, the remaining battery power of the terminal must be greater than a certain threshold, and the available communication bandwidth cannot exceed the channel capacity.

[0115] The cross-stage resource scheduling optimization function is a mathematical model that comprehensively considers the optimization objective, various constraints, and is used to solve for the optimal parameter configuration of multiple stages, including sensor sampling, data communication, and local processing. This function can be a linear programming, nonlinear programming, or mixed-integer programming model. Jointly solving for the sampling period, communication channel, and compression algorithm refers to simultaneously determining the sensor data acquisition frequency, the channel parameters used for data transmission, and the data compression intensity within the same optimization framework. This joint solution achieves global optima rather than local optimization. For example, heuristic or exact algorithms can be used for solving this problem.

[0116] The following is a concrete example to illustrate this. When constructing a cross-stage resource scheduling optimization function, the following steps can be followed. First, for the optimization objective, the semantic value density (e.g., a score from 0 to 100) can be multiplied by the expected transmission success rate of the data packet (e.g., a probability value from 0 to 1) to obtain the expected value utility. For example, if the semantic value density is 80 and the expected transmission success rate is 0.95, then the expected value utility is 76. This expected value utility is the objective that the optimization function needs to maximize. Second, when determining the maximum allowable reporting delay constraint, multiple water level intervals and their corresponding transmission urgency levels can be preset. For example, when the high confidence water level value is in the "normal water level interval," the transmission urgency level is "low," and the maximum allowable reporting delay constraint is set to 300 seconds. When the high confidence water level value is in the "warning water level interval," the transmission urgency level is "medium," and the maximum allowable reporting delay constraint is set to 60 seconds. When the high-confidence water level value falls within the "flood level range," the transmission urgency level is "high," and the maximum allowable reporting delay constraint is set to 10 seconds. The device dynamically selects the appropriate delay constraint based on the range the current high-confidence water level value falls into. Furthermore, when quantifying inequality constraints, the remaining energy of the terminal device (e.g., in milliampere-hours, mAh) and the available communication bandwidth of the current communication module (e.g., in Kbps) can be monitored in real time. For example, the remaining energy of the terminal can be set to be greater than 1000mAh to ensure that the device can operate for at least 24 hours. The available communication bandwidth can be set to be greater than 10Kbps to ensure basic data transmission needs. These conditions will serve as hard constraints in the optimization function to ensure the physical feasibility of the scheduling strategy. Finally, the expected value utility constructed above is used as the objective function, and the maximum allowable reporting delay constraint and inequality constraints such as the remaining energy of the terminal and the available communication bandwidth are incorporated into the optimization model. The optimization model can be implemented using mixed-integer linear programming (MILP), where decision variables include the sampling period (e.g., discrete values ​​such as 10 seconds, 30 seconds, 60 seconds, etc.), the communication channel selection (e.g., LoRa, NB-IoT, 4G, etc.), and the data compression algorithm (e.g., different levels of lossless and lossy compression). By solving this MILP model, the optimal combination of sampling period, communication channel parameters, and compression algorithm that maximizes the expected value utility while satisfying all constraints under the current conditions can be obtained.

[0117] The above implementation methods significantly improve the data reporting efficiency and decision support capabilities of intelligent water level monitoring devices in resource-constrained environments. Specifically, by constructing semantic value density and expected data packet transmission success rate as optimization objectives, the device ensures that it prioritizes and reliably transmits data of the highest value for hydrological event judgment and early warning, thus avoiding resource waste and omission of key information due to a lack of value orientation in traditional solutions. Simultaneously, by dynamically determining and quantifying the transmission urgency level based on high-reliability water level values ​​as the maximum allowable reporting delay constraint, the device can flexibly adjust the timeliness requirements of data reporting according to real-time changes in the hydrological situation, effectively responding to sudden hydrological events and improving the response speed and accuracy of early warnings. Furthermore, by quantifying the actual resource status, such as remaining terminal energy and available communication bandwidth, as inequality constraints, the generated scheduling strategy ensures that while meeting data value and timeliness requirements, it does not exceed the physical capacity of the equipment, thereby extending the equipment's runtime and guaranteeing long-term stable operation. Ultimately, by constructing a cross-stage resource scheduling optimization function that integrates these elements, joint optimization of multiple stages such as sampling, communication, and processing was achieved, making resource allocation more accurate and efficient, and significantly improving the intelligence level of the entire monitoring device and its resilience in dealing with complex hydrological environments.

[0118] This application proposes a method for solving the cross-link resource scheduling optimization function and generating a segmented reporting strategy package. The method includes: employing a hybrid solution strategy based on branch-and-bound and a greedy heuristic, solving the cross-link resource scheduling optimization function within the time limit allowed by the current full-link resource state, obtaining the target frequency for sensor sampling, the target channel and power for data communication, and the target compression ratio for local processing. The target frequency, target channel and power, and target compression ratio are then converted into specific hardware control parameters executable by the dual-body sensor, multi-physics-related sensor, communication module, and processing module. These specific hardware control parameters and their corresponding execution trigger conditions are encapsulated into standardized instruction blocks, and multiple standardized instruction blocks are assembled according to their execution sequence to obtain the segmented reporting strategy package.

[0119] Hybrid solution strategies combine the advantages of different optimization algorithms, aiming to balance solution quality and computational efficiency. Branch and bound algorithms typically find the global optimum, but have high computational complexity. Greedy heuristics, on the other hand, find local optima relatively quickly. Combining the two can significantly shorten the solution time while maintaining a certain level of solution quality, making it particularly suitable for resource-constrained scenarios with real-time requirements. For example, a greedy heuristic can be used to quickly obtain an initial feasible solution, which can then be used as the initial upper or lower bound for the branch and bound algorithm, thereby pruning the search space and accelerating the branch and bound process. Alternatively, at each node of the branch and bound algorithm, a greedy heuristic can be used to quickly evaluate the feasibility or potential optimality of subproblems to determine whether to continue branching or prune. The solution time limit refers to the maximum time that a device can tolerate in completing the optimization problem solution. In edge computing environments, computational and time resources are limited. Setting a solution time limit ensures that the optimization process does not occupy resources indefinitely, thus affecting the device's real-time response capability and overall performance. For example, the solution time limit can be dynamically adjusted based on real-time resource status such as the current device's CPU load, memory usage, network bandwidth, and expected task deadline. Alternatively, the solution time limit can be pre-configured as a fixed value or set in stages according to the urgency and semantic value density of the hydrological event. The target frequency of sensor sampling, the target channel and power of data communication, and the target compression ratio of local processing are key optimization scheduling parameters obtained after solving the cross-stage resource scheduling optimization function. They directly determine the specific behavior of sensor data acquisition, transmission, and local processing. The target frequency refers to the number of times the sensor collects data per unit time, affecting data freshness and energy consumption. The target channel and power refer to the communication path and transmission power selected during data transmission, affecting transmission reliability, latency, and energy consumption. The target compression ratio refers to the degree of data compression during local processing, affecting the amount of data transmitted, processing energy consumption, and data accuracy. These parameters can be expressed in numerical form, for example, a target frequency of 10 times per minute, a target channel of a LoRaWAN frequency point, a power of 14dBm, and a compression ratio of 50%. Alternatively, these parameters can be discrete levels or modes, such as target frequency in three levels: "high", "medium", and "low", target channel in options such as "cellular network", "LoRaWAN", and "satellite communication", and target compression rate in options such as "no compression", "light compression", and "deep compression".

[0120] The specific hardware control parameters that can be executed by dual-body sensors, multi-physics-field co-sensors, communication modules, and processing modules are instructions that can be directly recognized and executed by hardware devices after the optimization scheduling parameters have been transformed. Transforming abstract optimization parameters into specific hardware instructions is a crucial step in policy deployment, ensuring that the optimization results are accurately applied to physical devices. For example, for sensors, specific hardware control parameters can be register configuration values ​​used to set sampling interval registers, ADC sampling rate registers, etc. For communication modules, these can be AT command sets or SPI / I2C configuration commands used to set transmit power, channel frequency, modulation method, etc. For processing modules, this can be calling specific compression library functions and passing in compression level parameters. A standardized instruction block is a structured data unit that binds hardware control parameters with the conditions under which these parameters are executed. This encapsulation improves the manageability, reusability, and portability of instructions, making the generation and parsing of policy packages more efficient and reliable. For example, a standardized instruction block can be a data structure or object containing fields such as command_id, target_device, parameter_list, trigger_condition, and priority. Alternatively, it can be a message format defined based on a specific protocol, which includes instruction types, parameter payloads, and triggering rules. Arranging multiple standardized instruction blocks according to a predetermined temporal or logical order forms a complete execution sequence. This ensures that complex scheduling strategies are implemented step-by-step according to the design intent; for example, adjusting the sampling frequency first, then switching communication channels, and finally compressing data. This can be implemented, for example, through an instruction sequence list or queue, where each element is a standardized instruction block with a timestamp or execution order index. Alternatively, a state machine or flowchart can be used to represent the temporal dependencies between instruction blocks, ensuring that subsequent instructions are executed only after preconditions are met. The segmented reporting strategy package is the final, deployable set of instructions for edge devices. It contains optimized scheduling instructions for different stages (sampling, communication, processing) and is organized according to execution timing. This strategy package is the core carrier for achieving dynamic, intelligent resource scheduling. For example, the segmented reporting strategy package can be a binary file containing an encoded sequence of standardized instruction blocks and metadata. Alternatively, it can be a text file that describes the various components and execution logic of the strategy in a structured manner.

[0121] The following is a concrete example to illustrate this. Suppose that at a smart water level monitoring node, the data reporting strategy needs to be dynamically adjusted based on the current hydrological event semantics and resource status. First, when solving the cross-link resource scheduling optimization function, the edge device (e.g., the main control unit equipped with an ARM Cortex-M series microcontroller) will adopt a hybrid solution strategy based on branch and bound and greedy heuristics. For example, a greedy heuristic search based on a genetic algorithm can be run first to quickly find a near-optimal combination of sampling frequency, communication channel, and data compression ratio. Then, this combination is used as the initial solution of the branch and bound algorithm, and a solution time limit is set, such as 500 milliseconds. The branch and bound algorithm performs a fine search based on this, avoiding invalid calculations through pruning operations, ensuring that an optimized solution that meets the accuracy requirements is output within 500 milliseconds. Suppose the solution result is: the target sampling frequency of the sensor is 15 times per minute, the target channel of data communication is the EU868 band of LoRaWAN, the transmit power is 10dBm, and the target compression ratio of the local processing is the JPEG 2000 algorithm with a quality factor of 70. Next, these optimized scheduling parameters are translated into specific hardware control parameters. For example, for dual-body sensors and multi-physics-related sensors, a target frequency of 15 times / minute is translated into setting its internal timer register to trigger data acquisition every 4 seconds. For communication modules (e.g., LoRa modules based on the Semtech SX1276 chip), the target channel EU868 band and 10dBm transmit power are translated into specific SPI command sequences to configure the chip's frequency register and power amplifier gain. For processing modules (e.g., the DSP core inside the microcontroller), a target compression ratio of 70 is translated into calling a pre-built JPEG 2000 compression library function and passing in the corresponding quality parameters. Subsequently, these specific hardware control parameters and their corresponding execution trigger conditions are encapsulated into standardized instruction blocks. For example, one instruction block might contain "Execute high-frequency sampling configuration (timer register value X) when the water level exceeds the warning line." Another instruction block might contain "Switch to the LoRaWAN low-power channel (channel configuration Y, power configuration Z) when network congestion exceeds the threshold." Another instruction block might contain the instruction "Compress image data by 70% quality before sending data packets (call compression function A, parameter 70)". These instruction blocks can be encapsulated in JSON format, with each JSON object containing an instruction ID, target device type, parameter list, and trigger condition. Finally, these standardized instruction blocks are assembled according to their execution sequence to form a segmented reporting strategy package. For example, the strategy package might first contain an "Initialize sensor" instruction block, then an "Adjust sampling frequency according to water level change rate" instruction block, followed by an "Select communication channel according to network conditions" instruction block, and finally a "Compress data and send" instruction block.These instruction blocks are arranged in a logical order to ensure that, during actual operation, the device can efficiently and accurately execute various tasks according to the preset optimization strategy.

[0122] The above implementation effectively solves the problem of insufficient real-time performance caused by excessively long optimization function solution time on resource-constrained edge devices. Specifically, a hybrid solution strategy based on branch-and-bound and greedy heuristics is adopted, enabling the device to quickly obtain high-quality resource scheduling optimization parameters within limited computing resources and strict response time limits. This efficient solution capability avoids decision delays caused by complex optimization calculations, ensuring that the intelligent water level monitoring device can respond promptly to constantly changing water conditions and resource status. Furthermore, abstract optimization parameters are converted into specific hardware control parameters executable by dual-body sensors, multi-physics-related sensors, communication modules, and processing modules, and further encapsulated into standardized instruction blocks and assembled into segmented reporting strategy packages in a time sequence, greatly simplifying the deployment and execution of the strategy. This not only improves the accuracy and reliability of the strategy but also reduces the complexity of device integration. Combined with the overall framework proposed in the above scheme for generating segmented reporting strategy packages based on semantic value density, high-confidence water level values, and the current full-link resource status, the implementation of this scheme enables the entire intelligent water level monitoring device to achieve seamless integration from intelligent decision-making to efficient execution. The device can dynamically adjust sensor sampling frequency, communication channel and power, and data compression intensity based on the potential value and urgency of hydrological events and currently available resources, thereby optimizing resource consumption while maximizing data value. This capability significantly improves the timeliness, accuracy, and transmission efficiency of monitoring data, especially during sudden hydrological events, ensuring timely reporting of critical information and providing strong support for decision-making in flood control and drought relief, thus comprehensively enhancing the device's early warning foresight and reliability.

[0123] This application proposes a method for executing segmented reporting strategy packages and collecting strategy execution performance data, associating and mapping the performance data with semantic types to generate a dual-loop evolutionary feedback signal. This method simultaneously collects actual energy consumption, network transmission latency, and event data reporting success rate during the execution of the segmented reporting strategy package, and calculates the value achievement rate based on the semantic value density corresponding to the reported data packet and the actual reporting success rate. Actual energy consumption refers to the electrical energy consumed by the sensor node or related processing and communication modules during the execution of the segmented reporting strategy package. This energy consumption can be collected by real-time monitoring of current and voltage using a power metering chip integrated into the device and performing integral calculations, or by estimating using a software estimation model combined with the device's operating mode and duration. Network transmission latency refers to the time required for data packets to be sent from the sensor node to the receiving end. This can be calculated by embedding a timestamp in the data packet and recording the arrival time at the receiving end, or by obtaining the round-trip time measurement function provided by the network protocol stack. The event data reporting success rate refers to the proportion of event data packets successfully sent and acknowledged by the receiving end within a specific time period, out of the total number of data packets sent. This can be determined by introducing an acknowledgment mechanism into the communication protocol and statistically analyzing the reception of acknowledgment messages, or by statistically analyzing the integrity verification of data packets and the results of business logic processing at the application layer. Semantic value density, obtained from preceding steps, represents the dynamic ratio of the decision importance of the current hydrological event to the resource consumption cost. Value achievement rate is a comprehensive indicator used to measure whether the expected semantic value density has been effectively achieved under actual resource consumption and reporting success rate. Its calculation can be achieved by multiplying or weighting the actual reporting success rate with the semantic value density, or by evaluating it through a more complex utility function.

[0124] Subsequently, the collected actual energy consumption, network transmission latency, event data reporting success rate, and value achievement rate are associated and encapsulated with the semantic type that triggered the current policy execution and the scheduling instruction parameters in the segmented reporting policy package to form a structured experience data unit. Association and encapsulation refers to logically organizing data items from different sources and of different types together to form a data record with a clear structure. This can be done by defining a data structure (such as a JSON object, XML document, or database table record) to store each data item as a field, or by adding metadata tags to the header or footer of the data packet to bind relevant information. A structured experience data unit is a complete, self-describing data record that includes policy execution performance indicators, the event context that triggered the policy, and the policy's own configuration.

[0125] Based on these structured empirical data units, the device generates both local evolutionary signals and cloud-based co-evolutionary signals in parallel, forming a dual-loop evolutionary feedback signal. Parallel generation means that the generation of local and cloud-based co-evolutionary signals can occur simultaneously without interference. This can be achieved by setting two independent processor threads or tasks on the device side, each responsible for generating one type of signal, or by copying the structured empirical data units and sending them to different processing modules for signal generation. Local evolutionary signals are feedback information used to guide the local device in self-adjustment and optimization. Cloud-based co-evolutionary signals are feedback information that is aggregated in the cloud to participate in global model training and collaborative optimization. The dual-loop evolutionary feedback signal is a hierarchical and collaborative feedback mechanism composed of both local and cloud-based evolutionary signals. Local evolutionary signals are used to directly update the local resource prediction model. The local resource prediction model is a model deployed on sensor nodes or edge devices to predict future resource consumption or system performance. This model can be a statistical model based on historical data, such as a linear regression model or a time series prediction model, or a lightweight machine learning model, such as a decision tree or a small neural network. Direct updates mean that local evolutionary signals on the device can be directly used as input to adjust or retrain the parameters of the local resource estimation model. Meanwhile, cloud-based co-evolutionary signals, after being anonymized, are used in cloud-based federated learning. Anonymization refers to operations such as anonymization, generalization, and encryption of data to eliminate or reduce sensitive information contained in the data, protecting user privacy and data security. Cloud-based federated learning is a distributed machine learning paradigm that allows multiple devices or organizations to collaboratively train a global model without sharing the original data.

[0126] As a specific implementation method, the above approach can be achieved as follows: On a smart water level monitoring node, when a segmented reporting strategy package is executed, the node activates a performance monitoring module. This module integrates a miniature power meter to monitor the instantaneous current and voltage of the sensor, communication module, and microcontroller during data acquisition, processing, and transmission in real time, and accumulates and calculates the actual energy consumption. Simultaneously, the communication module embeds a high-precision timestamp when sending data packets and records the round-trip time upon receiving confirmation from the cloud, thereby calculating the network transmission latency. The event data reporting success rate is obtained by statistically analyzing the ratio of the number of data packets successfully receiving cloud confirmation to the total number of data packets sent within a certain time window. For example, if 100 data packets are sent and 95 confirmations are received, the success rate is 95%. Furthermore, the node extracts the corresponding semantic value density from the metadata of the reported data packets and, combined with the actual reporting success rate, calculates the value achievement rate of this strategy execution using a preset utility function (e.g., value achievement rate = semantic value density * actual reporting success rate). Subsequently, the collected actual energy consumption, network transmission latency, event data reporting success rate, and value achievement rate are encapsulated into a structured data unit. This data unit can be a JSON-formatted string containing fields such as "energy_consumption", "latency", "success_rate", and "value_achievement", as well as the semantic type that triggered the current policy execution (e.g., "flood_warning", "normal_flow") and the specific scheduling instruction parameters in the segmented reporting policy package (e.g., "sampling_freq":"1Hz", "channel":LoRa_CH3", "compression_ratio":"0.5"). Based on this structured empirical data unit, the node generates two types of feedback signals in parallel. The device-local evolution signal can be directly used to update the node's internal local... Local resource estimation models. For example, if the local resource estimation model is a linear regression model based on historical data, this signal can be used as a new training sample. The model's weights can be fine-tuned using gradient descent to make its predictions of future energy consumption and latency more accurate. Simultaneously, the cloud-based co-evolutionary signal will anonymize this structured empirical data unit, for example, by removing device IDs and normalizing or aggregating values ​​such as energy consumption and latency, before sending the processed data to the cloud. In the cloud, these anonymized signals will converge and participate in a federated learning task to jointly train a global dynamic trust fusion weight generation model and a semantic value density calculation parameter model. For example, the cloud can aggregate model updates from multiple nodes to optimize the global sensor trust assessment algorithm, ensuring high accuracy across different environments and device types.

[0127] This application's solution effectively addresses the issues of insufficient utilization of performance data and lack of co-evolution in traditional water level monitoring devices through device-based performance data acquisition, tightly coupled encapsulation, and a dual-loop evolution mechanism, thereby significantly improving the overall device's adaptability and intelligence level. Specifically, during the execution of segmented reporting strategy packages, actual energy consumption, network transmission latency, and event data reporting success rate are collected simultaneously. The value achievement rate is calculated based on the semantic value density corresponding to the reported data packets and the actual reporting success rate. This step ensures a comprehensive and multi-dimensional quantification of the strategy execution effect. It not only covers key indicators such as resource consumption but also quantifies the actual benefits of data transmission through the value achievement rate, avoiding the one-sidedness of performance evaluation in traditional methods. This allows the device to accurately assess the actual value of strategy execution, providing a solid data foundation for subsequent optimization. The collected actual energy consumption, network transmission latency, event data reporting success rate, and value achievement rate are associated and encapsulated with the semantic type that triggered the current strategy execution and the scheduling instruction parameters in the segmented reporting strategy package to form structured experience data units. This process solves the problem of loose association between performance data and event semantics and strategy parameters by tightly binding the "result" and "cause" of strategy execution. These structured data units possess complete contextual information, greatly improving data utilization efficiency and facilitating efficient analysis and iterative optimization by the device, thereby enabling more accurate identification of the optimal scheduling strategy under specific hydrological events. Based on the structured experience data units, local evolutionary signals and cloud-based co-evolutionary signals are generated in parallel, forming a dual-loop evolutionary feedback signal. The local evolutionary signal is used to directly update the local resource prediction model, enabling rapid local adaptation of monitoring nodes. This allows each node to quickly adjust its resource usage strategy based on its real-time operating experience and resource status, improving the accuracy and response speed of local decision-making. Simultaneously, the cloud-based co-evolutionary signal, after de-sensitization processing, participates in cloud-based federated learning, introducing an efficient co-evolutionary mechanism and solving the problem of device fragmentation. Through federated learning, different monitoring nodes can share learning experiences while protecting data privacy, and jointly train a more robust and generalized global model, thereby promoting global knowledge sharing and continuous optimization, and systematically enhancing the ability to adapt to complex environmental changes.

[0128] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0129] This application also provides a schematic diagram of a smart water level monitoring device based on dual sensors and segmented reporting. The device includes: The fusion module is used to perform dynamic trust fusion of heterogeneous raw water level data collected by dual-body sensors and associated data collected by multi-physics associated sensors under the constraints of hydraulic mechanism, so as to obtain a high-confidence water level value and multi-physics coupled feature vector. This dynamic trust fusion is based on the joint judgment of sensor health evolution trend and simplified hydraulic consistency verification.

[0130] The analysis module is used to perform event semantic evolution analysis on the multi-physics coupled feature vector and the historical hydrological event sequence to obtain the semantic type, confidence level and semantic value density of the current hydrological event. The semantic value density is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost. The historical hydrological event sequence is derived from the recent high-value density data fragments cached locally. The generation module is used to generate a segmented reporting strategy package based on the semantic value density, the high confidence level value and the current full-link resource status. The segmented reporting strategy package contains a joint scheduling instruction for differentiated sampling frequency configuration, channel parameter selection and data compression intensity. The execution module is used to execute the segmented reporting strategy package and collect strategy execution performance data. It then associates and maps the performance data with the semantic type to generate a double-loop evolutionary feedback signal. This double-loop evolutionary feedback signal is used to iteratively update the weight generation rules of the dynamic trust fusion and the calculation parameters of the semantic value density.

[0131] It should be noted that the above embodiments of the intelligent water level monitoring device based on dual sensors and segmented reporting are only illustrated by the division of the functional modules described above. In practical applications, the functions described above can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the intelligent water level monitoring device based on dual sensors and segmented reporting provided in the above embodiments and the intelligent water level monitoring method embodiments based on dual sensors and segmented reporting belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0132] The above are merely optional embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A smart water level monitoring method based on dual sensors and segmented reporting, characterized in that, The method includes: Dynamic trust fusion under hydraulic mechanism constraints is performed on the heterogeneous raw water level data collected by dual-body sensors and the associated data collected by multi-physics associated sensors to obtain a high-confidence water level value and a multi-physics coupled feature vector. The multi-physics associated sensors are sensors used to collect other physical quantity data related to water level changes. The other physical quantity data includes flow velocity, turbidity, temperature or conductivity. The hydraulic mechanism constraints are the basic physical laws and principles used to describe the movement and behavior of water bodies. In data processing, they serve as verification rules to ensure that different physical quantity data conform to objective physical relationships. The dynamic trust fusion is a data fusion technology. Its core lies in dynamically adjusting the trust weight of each data source according to the real-time status of the sensors, data quality and consistency with the physical mechanism. The dynamic trust fusion is based on the joint judgment of the sensor health evolution trend and simplified hydraulic consistency verification. The event semantic evolution analysis is performed on the multiphysics field coupled feature vector and the historical hydrological event sequence to obtain the semantic type, confidence level and semantic value density of the current hydrological event. The semantic value density is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost. The historical hydrological event sequence is a set of hydrological event data fragments with high value density from the past period that are locally cached. Based on the semantic value density, the high confidence level value, and the current full-link resource status, a segmented reporting strategy package is generated. The segmented reporting strategy package includes a joint scheduling instruction for differentiated sampling frequency configuration, channel parameter selection, and data compression intensity. The segmented reporting strategy package is executed and strategy execution performance data is collected. The performance data is associated and mapped with the semantic type to generate a double-loop evolutionary feedback signal. The double-loop evolutionary feedback signal is used to iteratively update the weight generation rules of the dynamic trust fusion and the calculation parameters of the semantic value density.

2. The method according to claim 1, characterized in that, The process involves dynamically trusting the fusion of heterogeneous raw water level data acquired by the dual-body sensors and associated data acquired by the multi-physics associated sensors under hydraulic mechanism constraints to obtain a high-confidence water level value and a multi-physics coupled feature vector, including: Based on the instantaneous reading quality and historical reliability data of each sensor, the initial confidence weight of each sensor is determined. A simplified hydraulic relationship model is used to verify the physical consistency between the original heterogeneous water level data and the associated data, and a mechanism confidence factor is generated. Based on the initial confidence weight, the mechanism confidence factor, and the health index of each sensor, the target weight of each sensor in the fusion calculation is determined. The target weights are used to perform weighted calculations on the heterogeneous water level raw data of the dual-body sensors to generate the high-confidence water level value. The dual-body sensors are arranged in a preset geometric layout on the monitoring section. The associated data and the mechanism confidence factor are combined and correlated to obtain the multiphysics coupling feature vector.

3. The method according to claim 2, characterized in that, The process of using a simplified hydraulic relationship model to verify the physical consistency between the heterogeneous water level raw data and the associated data, and generating a mechanism confidence factor, includes: Based on the historical water level data sequence of the dual-body sensors, the current water level change rate is determined; Based on the simplified hydraulic relationship model, a theoretical associated parameter expected range that matches the current water level change rate and the current water level value is determined. The theoretical associated parameter includes at least one of the expected range of flow velocity or the expected range of turbidity. The actual associated parameters in the associated data collected by the multiphysics associated sensor are compared with the expected range of the theoretical associated parameters to obtain the parameter deviation. Based on the parameter deviation, a mechanism confidence factor is generated to characterize the physical consistency between the heterogeneous original water level data and the associated data. The greater the parameter deviation, the smaller the mechanism confidence factor.

4. The method according to claim 1, characterized in that, The step of performing event semantic evolution analysis on the multiphysics coupled feature vector and historical hydrological event sequences to obtain the semantic type, confidence level, and semantic value density of the current hydrological event includes: Feature analysis is performed on the multiphysics coupled feature vectors and combined with the historical hydrological event sequence to extract trend features that characterize the evolution of hydrological events; The trend features are matched and reasoned with a predefined hydrological event semantic rule base to obtain the semantic type and confidence level of the current hydrological event. The matching and reasoning process incorporates temporal context verification based on the historical hydrological event sequence. The semantic value density is determined based on the semantic type, the confidence level, and the estimated resource consumption cost.

5. The method according to claim 4, characterized in that, The feature analysis of the multiphysics coupled feature vector and the joint analysis of the historical hydrological event sequence are used to extract trend features characterizing the evolution of hydrological events, including: Construct an event state transition probability matrix based on the historical hydrological event sequence; The multiphysics coupling feature vector is dynamically time-warped and matched with the historical feature vector corresponding to the historical hydrological event sequence to obtain the similarity and phase difference between the current event and the historical event evolution stage. Based on the event state transition probability matrix, the similarity, and the phase difference, the trend probability vector of the current hydrological event evolving into each potential subsequent event type is determined, and the trend probability vector is used as the trend feature.

6. The method according to claim 4, characterized in that, Determining the semantic value density based on the semantic type, the confidence level, and the estimated resource consumption cost includes: Based on the preset decision importance weights corresponding to the semantic types and the confidence levels, dynamic event weights are calculated using a nonlinear correction function. Based on the historical resource consumption records in the historical hydrological event sequence, as well as the semantic type and the confidence level, the estimated resource consumption cost of transmitting data related to the current hydrological event is estimated. The dynamic event weights and the estimated resource consumption costs are input into the online optimizer to solve for the semantic value density. The online optimizer aims to maximize the utility function defined by the dynamic event weights and the estimated resource consumption costs within the current decision-making cycle.

7. The method according to claim 1, characterized in that, The step of generating a segmented reporting strategy package based on the semantic value density, the high-confidence water level value, and the current end-to-end resource status includes: Based on the semantic value density and referring to the strategy execution performance data recorded in the historical hydrological event sequence, a dynamic priority strategy for resource allocation is determined. With the goal of maximizing the expected value utility in the next decision cycle, and constrained by the water situation reflected by the high-confidence water level value and the current full-link resource status, a cross-link resource scheduling optimization function is constructed. Solve the cross-stage resource scheduling optimization function to obtain optimized scheduling parameters for sensor sampling, data communication and local processing stages, and generate the segmented reporting strategy package containing execution instructions based on the optimized scheduling parameters.

8. The method according to claim 7, characterized in that, The objective is to maximize the expected value utility in the next decision-making cycle. Constrained by the water situation reflected by the high-confidence water level and the current end-to-end resource status, a cross-link resource scheduling optimization function is constructed, including: The semantic value density is used as a value weight coefficient, and together with the expected transmission success rate of the data packet, the expected value utility is constructed as an optimization objective. Based on the water level range where the high-confidence water level value is located, the transmission urgency level corresponding to the water level range is determined, and the transmission urgency level is quantified as the maximum allowable reporting delay constraint. The remaining terminal energy and available communication bandwidth in the current full-link resource state are quantified into inequality constraints. By combining the optimization objective, the maximum allowable reporting delay constraint, and the inequality constraint, the cross-stage resource scheduling optimization function is constructed. This function is used to jointly solve for the sampling period, communication channel, and compression algorithm.

9. The method according to claim 1, characterized in that, The process of executing the segmented reporting strategy package and collecting strategy execution performance data, associating and mapping the performance data with the semantic type, and generating a double-loop evolutionary feedback signal includes: During the execution of the segmented reporting strategy package, actual energy consumption, network transmission latency, and event data reporting success rate are collected synchronously, and the value achievement rate is calculated based on the semantic value density corresponding to the reported data packet and the actual reporting success rate. The actual energy consumption, network transmission delay, event data reporting success rate, and value achievement rate collected are associated and encapsulated with the semantic type that triggered this strategy execution and the scheduling instruction parameters in the segmented reporting strategy package to form a structured experience data unit. Based on the structured empirical data unit, a device local evolution signal and a cloud-based co-evolution signal are generated in parallel to form the dual-loop evolutionary feedback signal. The device local evolution signal is used to directly update the local resource prediction model, and the cloud-based co-evolution signal is used to participate in cloud-based federated learning after being de-identified.

10. An intelligent water level monitoring device based on dual sensors and segmented reporting, characterized in that, The device includes: The fusion module is used to perform dynamic trust fusion of heterogeneous raw water level data collected by dual-body sensors and associated data collected by multi-physics associated sensors under hydraulic mechanism constraints, to obtain a high-confidence water level value and a multi-physics coupled feature vector. The multi-physics associated sensors are sensors used to collect other physical quantity data related to water level changes. The other physical quantity data includes flow velocity, turbidity, temperature or conductivity. The hydraulic mechanism constraints are the basic physical laws and principles used to describe the movement and behavior of water bodies. In data processing, they serve as verification rules to ensure that different physical quantity data conform to objective physical relationships. The dynamic trust fusion is a data fusion technology. Its core lies in dynamically adjusting the trust weight of each data source based on the real-time status of the sensors, data quality, and consistency with the physical mechanism. The dynamic trust fusion is based on the joint judgment of the sensor health evolution trend and simplified hydraulic consistency verification. The analysis module is used to perform event semantic evolution analysis on the multiphysics coupling feature vector and the historical hydrological event sequence to obtain the semantic type, confidence level and semantic value density of the current hydrological event. The semantic value density is determined by the dynamic ratio of the decision importance weight of the semantic type to the resource consumption cost. The historical hydrological event sequence is a set of hydrological event data fragments with high value density from the past period that are locally cached. The generation module is used to generate a segmented reporting strategy package based on the semantic value density, the high confidence level value and the current full-link resource status. The segmented reporting strategy package includes a joint scheduling instruction for differentiated sampling frequency configuration, channel parameter selection and data compression intensity. The execution module is used to execute the segmented reporting strategy package and collect strategy execution performance data, associate and map the performance data with the semantic type, and generate a double-loop evolutionary feedback signal. The double-loop evolutionary feedback signal is used to iteratively update the weight generation rules of the dynamic trust fusion and the calculation parameters of the semantic value density.

Images